Upload
others
View
3
Download
0
Embed Size (px)
Citation preview
UNIVERSITÉ DU QUÉBEC À RIMOUSKI
SIMULATION ET VALIDATION EXPÉRIMENTALE DU
TRAITEMENT THERMIQUE SUPERFICIEL AU LASER
APPLIQUÉ À DES GÉOMÉTRIES COMPLEXES
Mémoire présenté
dans le cadre du programme de maîtrise en ingénierie
en vue de l’obtention du grade de maître ès sciences appliquées (M.Sc.A.)
PAR
© GUILLAUME BILLAUD
mars 2016
ii
Composition du jury :
Éric Hudier, président du jury, Université du Québec À Rimouski
Abderazzak El Ouafi, directeur de recherche, Université du Québec À Rimouski
Nourredine Barka, codirecteur de recherche, Université du Québec À Rimouski
Souheil-Antoine Tahan, examinateur externe, École de Technologie Supérieure
Dépôt initial le 12 janvier 2016 Dépôt final le 15 mars 2016
UNIVERSITÉ DU QUÉBEC À RIMOUSKI
Service de la bibliothèque
Avertissement
La diffusion de ce mémoire ou de cette thèse se fait dans le respect des droits de son auteur,
qui a signé le formulaire « Autorisation de reproduire et de diffuser un rapport, un
mémoire ou une thèse ». En signant ce formulaire, l’auteur concède à l’Université du
Québec à Rimouski une licence non exclusive d’utilisation et de publication de la totalité
ou d’une partie importante de son travail de recherche pour des fins pédagogiques et non
commerciales. Plus précisément, l’auteur autorise l’Université du Québec à Rimouski à
reproduire, diffuser, prêter, distribuer ou vendre des copies de son travail de recherche à des
fins non commerciales sur quelque support que ce soit, y compris l’Internet. Cette licence
et cette autorisation n’entraînent pas une renonciation de la part de l’auteur à ses droits
moraux ni à ses droits de propriété intellectuelle. Sauf entente contraire, l’auteur conserve
la liberté de diffuser et de commercialiser ou non ce travail dont il possède un exemplaire.
v
REMERCIEMENTS
L’auteur de ce mémoire tient à remercier le professeur Abderrazak El Ouafi de
l’UQAR d’avoir dirigé mon mémoire, pour tous les conseils qu’il m’a donnés notamment
pour les vérifications expérimentales ainsi que pour le temps investi dans la correction de
mes travaux.
Également, l’auteur souhaite remercier l’étudiant de l’UQAR, Al-Khader Borki pour
sa contribution à la recherche, notamment par la fabrication du banc d’essais ayant servi à
la vérification expérimentale ainsi que pour son assistance et ses conseils.
vii
RÉSUMÉ
Le traitement thermique superficiel au laser est un procédé qui vise à améliorer la
résistance des pièces mécaniques à l’usure et à la fatigue en augmentant la dureté des
surfaces critiques. Se distinguant par son apport thermique bref et localisé, par sa capacité
en matière de puissance surfacique et par ses cycles thermiques rapides et précis, ce
procédé permet d’améliorer la résistance à l’usure et à la fatigue en durcissant les zones
critiques superficielles de la pièce tout en limitant les risques de déformations indésirables.
Les caractéristiques mécaniques de la zone durcie obtenue par traitement thermique au laser
dépendent des propriétés physicochimiques du matériau à traiter et de plusieurs paramètres
du procédé lui-même. L’exploitation adéquate des possibilités qu’offre ce procédé nécessite
le développement de stratégies permettant de contrôler ces paramètres de manière à
produire avec précision les caractéristiques désirées sans recourir au traditionnel long et
couteux processus essai-erreur. L’objectif du projet consiste à analyser les relations de
dépendance entre le profil de dureté et les paramètres du procédé dans le but d’établir des
modèles simples permettant la prédiction du profil de dureté dans le cas de traitement de
pièces mécaniques en acier AISI 4340 de géométries complexes. Une grande partie de ces
analyses a été réalisée grâce à des simulations sur des modèles numériques 3D utilisant la
méthode des éléments finis. Pour arriver à des modèles prédictifs consistants, une approche
en trois phases a été adoptée. La première a consisté à développer le modèle 3D et à le
valider expérimentalement pour des géométries simples. Cette phase a permis d’analyser
les effets des différents paramètres sur le profil de dureté en se basant sur une
expérimentation structurée combinée à des techniques éprouvées d’analyse statistique. Les
résultats de cette étude ont conduit à l’identification des variables les plus pertinentes à
exploiter pour la modélisation. La seconde phase a permis d’enrichir le modèle 3D et les
résultats de simulation pour les combiner avec les données expérimentales dans le but
d’élaborer les modèles prédictifs les plus précis possible. Deux techniques de modélisation
ont été considérées à cet effet, soient la régression multiple et les réseaux de neurones.
Enfin, la troisième phase a été consacrée au développement du modèle 3D pour des
géométries complexes et à sa validation expérimentale. Ce modèle a été appliqué avec
succès dans le cas d’un engrenage. Au cours des trois phases, les résultats obtenus se sont
avérés très satisfaisants et ont montré une concordance remarquable entre les prédictions et
les mesures expérimentales.
Mots clés: traitement thermique par laser, traitement thermique superficiel,
simulation 3D, profile de dureté, réseau de neurones, engrenage
ix
ABSTRACT
Laser surface hardening transformation is a superficial heating process which aims to
enhance the wear and fatigue resistance by hardening the superficial critical areas of
mechanical parts. This process is well-known by his capacity in terms of power flux density
and recognized by his fast, local and accurate thermal cycles, while limiting the risks of
undesirable distortion and deformation effects. The mechanical properties of the hardened
surface depend of the physicochemical properties of the material as well as the heating
system parameters. To adequately exploit the advantages presented by this heating method,
it is necessary to develop a comprehensive strategy to control and adjust the process
parameters in order to produce desired hardened surface characteristics without being
forced to use the traditional and fastidious trial and error procedures. This study aims to
analyse the relationship between the hardness profile and the process parameters in order to
build a basic prediction models for hardness profile for AISI 4340 steel mechanical parts
with complex geometries. The presented results in the study were achieved using finite
elements method based numerical 3D models. To reach accurate and robust hardness
profile predictive models, a three-step approach was adopted. The first step consists to
develop and test experimentally the numerical 3D model for simple geometries. The model
is used to evaluate the effects of different parameters on the hardness profile using a
structured experimental design combined to confirmed statistical analysis tools. The results
of this phase permitted the identification of the most relevant variables to use in the
modeling stage. Then, the second step consists of enhancing and enriching the numerical
model used to generate the modelling data base in order to improve the accuracy and the
robustness of the predictive models. Two modeling techniques have been considered for the
modelling purpose: multiple regression and neural networks. Finally, the third step consists
of adapting various generated models to complex geometries. The models was successfully
applied and experimentally confirmed in the case of a gear. The results obtained in each
step were very satisfying and showed great concordance between predicted results and
experimental measures.
Keywords: laser heat treatment, surface transformation hardening, 3D simulation,
case depth, hardness profile, neural network, gears
xi
TABLE DES MATIÈRES
REMERCIEMENTS ........................................................................................................... vi
RÉSUMÉ ............................................................................................................................ viii
ABSTRACT .......................................................................................................................... x
TABLE DES MATIÈRES ................................................................................................. xii
LISTE DES TABLEAUX .................................................................................................. xv
LISTE DES FIGURES.................................................................................................... xviii
INTRODUCTION GÉNÉRALE ........................................................................................ 1
0.1-LE PRINCIPE DU LASER .................................................................................................. 1
0.2-LE TRAITEMENT THERMIQUE DES ACIERS ................................................................... 2
0.3- LE TRAITEMENT THERMIQUE AU LASER ...................................................................... 7
0.4- AVANTAGES ET INCONVÉNIENTS DU PROCÉDÉ DE TRAITEMENT THERMIQUE
SURFACIQUE AU LASER. ........................................................................................... 10
0.5-PROBLÉMATIQUE GÉNÉRALE ...................................................................................... 12
0.6-OBJECTIF ..................................................................................................................... 13
0.7-MÉTHODOLOGIE ......................................................................................................... 14
0.8-ORGANISATION DU MÉMOIRE ..................................................................................... 14
CHAPITRE 1 PRÉDICTION DU PROFIL DE DURETÉ D’UNE PLAQUE EN
ACIER 4340 TRAITÉE THERMIQUEMENT AU LASER EN UTILISANT UN
MODEL 3D ET UNE VALIDATION EXPÉRIMENTALE .......................................... 16
1.1-RESUME EN FRANÇAIS DU PREMIER ARTICLE ............................................................ 16
1.2-PREDICTION OF HARDNESS PROFILE OF 4340 STEEL PLATE HEAT TREATED BY
LASER USING 3D MODEL AND EXPERIMENTAL VALIDATION ................................... 18
xiii
CHAPITRE 2 MODÈLE DE RÉSEAU DE NEURONES ARTICFICIELS
POUR L’ESTIMATION DU TRAITEMENT THERMIQUE SUPERFICIEL
AU LASER D’UNE PLAQUE D’ACIER 4340 ................................................................ 41
2.1-RESUME EN FRANÇAIS DU DEUXIEME ARTICLE .......................................................... 41
2.2- ANN BASED MODEL FOR ESTIMATION OF TRANSFORMATION HARDENING OF
AISI 4340 STEEL PLATE HEAT-TREATED BY LASER ................................................ 43
CHAPITRE 3 ANALYSE THERMIQUE DU TRAITEMENT SURFACIQUE
AU LASER DES ENGRENAGES-SIMULATION 3D ET VALIDATION .................. 73
3.1-RESUME EN FRANÇAIS DU TROISIEME ARTICLE .......................................................... 73
3.2- THERMAL ANALYSIS OF SURFACE TRANSFORMATION HARDENING OF GEARS
USING LASER – 3D SIMULATION AND VALIDATION .................................................. 75
CONCLUSION GÉNÉRALE ............................................................................................ 99
RÉFÉRENCES BIBLIOGRAPHIQUES ....................................................................... 102
xiv
LISTE DES TABLEAUX
Tableau 0.1 : Comparaison entre les différents procédés ..................................................... 11
Tableau 1.1 : Material properties .......................................................................................... 27
Tableau 1.2 : Experimental matrix for validation ................................................................. 34
Tableau 1.3 : Average absolute and relative hardness errors ............................................... 38
Tableau 1.4 : Comparison of hardened depth ....................................................................... 39
Tableau 2.1 : Metallurgical properties .................................................................................. 49
Tableau 2.2 : Experimental matrix for validation ................................................................. 53
Tableau 2.3 : Average absolute and relative hardness errors resulting from the
preliminary tests .................................................................................................................... 54
Tableau 2.4 : Corrected Rc according to process parameters ............................................... 56
Tableau 2.5 : Factors and levels used for the ANOVA study ............................................... 58
Tableau 2.6 : ANOVA for HD .............................................................................................. 59
Tableau 2.7 : Results of the ANOVA for HBW ................................................................... 61
Tableau 2.8 : Middle points .................................................................................................. 66
Tableau 2.9 : Comparison of the results ............................................................................... 69
Tableau 2.10 : Experimental matrix for validation ............................................................... 70
Tableau 2.11 : Experimental validation – results ................................................................. 71
Tableau 3.1 : : Material properties ........................................................................................ 81
Tableau 3.2 : Experimental matrix ....................................................................................... 91
xvi
Tableau 3.3 : Maximum absolute and relative hardness errors ............................................ 94
Tableau 3.4 : Comparison between simulation and experiment for T0 = 873 K .................. 97
xvii
LISTE DES FIGURES
Figure 0.1 : Amplification de photons. Principe du laser ....................................................... 2
Figure 0.2 : diagramme de phase fer-carbone ........................................................................ 3
Figure 0.3 : perlite observée au microscope ........................................................................... 5
Figure 0.4 : Bainite observée au microscope ......................................................................... 6
Figure 0.5 : Structure martensitique ....................................................................................... 7
Figure 0.6 : Principe du traitement surfacique par laser ........................................................ 8
Figure 0.7 : Profils de dureté pour différentes vitesses de balayage dans le cas d’un
acier AISI 4340 ...................................................................................................................... 9
Figure 1.1 : Transformation from perlite into austenite ....................................................... 24
Figure 1.2 : Homogenization of hypoeutectoid steel ........................................................... 24
Figure 1.3 : Final used mesh (dimensions are in mm) ......................................................... 28
Figure 1.4 : Effect of mesh size on obtained temperatures .................................................. 29
Figure 1.5 : Isothermal contours of temperature (t = 1 s) .................................................... 30
Figure 1.6 : Isothermal contours of temperature (t = 2 s) .................................................... 31
Figure 1.7 : Isothermal contours of temperature (t = 3 s) .................................................... 31
Figure 1.8 : Temperature versus time for various depths measured from heated
surface at middle plan .......................................................................................................... 32
Figure 1.9 : Temperature profile along the y-axis at a given time and at different
depths ................................................................................................................................... 33
Figure 1.10 : Experimental setup ......................................................................................... 34
xix
Figure 1.11 : Metallographic picture of hardness profile - Test ........................................... 35
Figure 1.12 : Hardness curve for test 1 (950 W and 16 mm/s) ............................................. 36
Figure 1.13 : Hardness curve for test 1 (950 W and 18 mm/s) ............................................. 37
Figure 1.14 : Hardness curve for test 1 (1100 W and 16 mm/s) ........................................... 37
Figure 1.15 : Hardness curve for test 1 (1100 W and 18 mm/s) ........................................... 38
Figure 2.1 : Sample with its mesh implemented on COMSOL ............................................ 47
Figure 2.2 : Isothermal contours: (a) --» t = 0.4 s and (b) --» t = 2.5 s ................................. 49
Figure 2.3 : Temperature evolution at different depths (850 W and 9 mm/s) ...................... 51
Figure 2.4 : Experimental setup for model validation .......................................................... 53
Figure 2.5 : Hardness curve for test 1 (850 W and 9 mm/s) ................................................. 54
Figure 2.6 : Hardness curve for test 1 (850 W and 12 mm/s) ............................................... 55
Figure 2.7 : Hardness curve for test 1 (950 W and 12 mm/s) ............................................... 55
Figure 2.8 : Micrographic picture illustrating the HBW and HD after chemical
etching ................................................................................................................................... 57
Figure 2.9 : Effects of parameters on case depth .................................................................. 60
Figure 2.10 : Comparison between simulated HD and HD calculated by regression
formula (Equation (12)) ........................................................................................................ 61
Figure 2.11 : Main effects plot for hardened bead width ...................................................... 62
Figure 2.12 : Comparison between simulated HBW and HBW calculated by
regression formula (Equation (13)) ....................................................................................... 63
Figure 2.13 : Principle of the neural network ....................................................................... 65
Figure 2.14 : Absolute relative errors for HD and HBW ...................................................... 67
Figure 2.15 : Relative errors for HD and HBW .................................................................... 67
xx
Figure 2.16 : Comparison between simulated HD and HD calculated by the neural
network ................................................................................................................................. 68
Figure 2.17 : Comparison between simulated HBW and HBW calculated by the
neural network ...................................................................................................................... 69
Figure 3.1 : Laser trajectory ................................................................................................. 80
Figure 3.2 : The gear mounted on the shaft and clamped by two mounting rings ............... 80
Figure 3.3 : Isothermal contours for P = 2500 W, Vsc = 1 mm/s, wr = 240 rpm, T0 =
873 K .................................................................................................................................... 82
Figure 3.4 : Isothermal contours for P = 1500 W, Vsc = 1 mm/s, wr= 60 rpm, T0 = 500
K ........................................................................................................................................... 83
Figure 3.5 : Isothermal contours for P = 1500 W, Vsc = 1 mm/s, wr= 120 rpm, T0 =
500 K .................................................................................................................................... 84
Figure 3.6 : : Isothermal contours for P = 1500 W, Vsc =1 mm/s, wr= 300 rpm, T0 =
500 K .................................................................................................................................... 84
Figure 3.7 : Experimental setup ........................................................................................... 87
Figure 3.8 : Case depth observation for P = 3000 W, Vsc = 0.75 mm/s, wr = 300 rpm
and T0 = 500 K ..................................................................................................................... 89
Figure 3.9 : Case depth observation for P = 3000 W, Vsc = 0.50 mm/s and wr= 300
rpm and T0 = 500 K .............................................................................................................. 90
Figure 3.10 : Case depth observation for P = 2500 W, Vsc = 1 mm/s and wr= 240 rpm
and T0 = 600 K ..................................................................................................................... 91
Figure 3.11 : Hardness curve for P = 2500 W, Vsc = 0.75 mm/s and wr= 240 rpm and
T0 = 600 K ............................................................................................................................ 92
Figure 3.12 : Hardness curves for P = 2500 W, Vsc = 0.75 mm/s and wr= 300 rpm and
T0 = 600 K ............................................................................................................................ 93
Figure 3.13 : Hardness curves for P = 3000 W, Vsc = 0.75 mm/s and wr = 300 rpm
and T0 = 600 K ..................................................................................................................... 94
xxi
Figure 3.14 : Case depth observation for P= 1500W, Vsc = 0.75 mm/s and wr= 750
rpm and T0 = 873 K .............................................................................................................. 95
Figure 3.15 : Hardness curves for P= 1500 W, Vsc = 0.75 mm/s and wr= 750 rpm and
T0 = 873 K at the top of the tooth ......................................................................................... 96
Figure 3.16 : Hardness curves for P = 1500W, Vsc = 0.75 mm/s and wr= 750 rpm and
T0 = 873 K at the foot of the tooth ........................................................................................ 97
INTRODUCTION GÉNÉRALE
De nos jours, le traitement thermique superficiel au laser est un procédé de plus en
plus utilisé dans les industries automobiles et aéronautiques. Ce procédé vise à donner, au
cœur et à la surface, des pièces mécaniques des caractéristiques différentes permettant
d’améliorer leur résistance à l’usure et à la fatigue en durcissant les zones critiques
superficielles tout en limitant les risques de déformations indésirables. Son avantage réside
dans son apport thermique bref et localisé, sa capacité en termes de puissance surfacique et
ses cycles thermiques rapides et précis. Les caractéristiques mécaniques de la zone durcie
obtenue par traitement thermique au laser dépendent des propriétés physicochimiques du
matériau à traiter et de plusieurs paramètres du procédé lui-même. Ce mémoire traite de la
simulation et de la validation expérimentale du traitement thermique superficiel au laser des
pièces mécaniques en acier AISI 4340 de géométries complexes telles que les engrenages.
Il vise à analyser les effets des paramètres du procédé sur le profil de dureté dans le but
d’élaborer des modèles prédictifs de la qualité du traitement thermique.
0.1-LE PRINCIPE DU LASER
Le laser (acronyme de l’anglais « Light Amplification by Stimulated Emission of
Radiation ») se présente généralement sous la forme d’un dispositif qui émet un
rayonnement cohérent sur le plan spatial et temporel. Le principe du laser repose sur trois
phénomènes physiques décrivant l’interaction entre l’atome et la lumière. Le premier
principe est celui de l’absorption : un atome peut recevoir et absorber un photon possédant
une longueur d’onde bien définie. Il passe alors dans un état dit « excité ». Le deuxième
principe est celui de l’émission spontanée : un atome excité peut revenir à son état initial
(dit « état fondamental ») en émettant spontanément un photon de même longueur d’onde
que celui qu’il avait absorbé. La direction et la phase du photon émis spontanément sont
2
aléatoires. Le troisième et dernier principe est celui de l’émission stimulée. Un atome excité
qui reçoit un photon qui aurait permis de l’exciter s’il était dans son état fondamental peut
se désexciter en émettant un photon dans la même direction et avec la même phase que le
photon incident. Ce photon émis va alors s’ajouter au rayonnement et l’amplifier [1-3]. La
Figure 0.1 ci-après montre le principe d’amplification par émission stimulée.
Photon 2 photons
Émission stimulée
Amplification par émission stimulée
atome
Figure 0.1 : Amplification de photons. Principe du laser
Il existe plusieurs types de laser qui se différencient par leur milieu amplificateur.
Ainsi, il existe le laser à hélium néon, à argon, au krypton, au CO2…
0.2-LE TRAITEMENT THERMIQUE DES ACIERS
L’acier est un alliage fer-carbone quelquefois combiné à d’autres éléments tels que le
manganèse, le nickel, le chrome, etc. L'acier présente plusieurs phases solides (ferrite,
perlite, cémentite, austénite) avec des températures de changement de phase ou d’états
variables selon la teneur en carbone. La Figure 0.2 montre les frontières entre les
3
différentes phases et états de l’acier. En abscisse, nous avons la teneur en carbone et en
ordonnée, la température.
Carbon Weight Percent
Te
mp
era
ture
in C
els
ius
Figure 0.2 : Diagramme de phase fer-carbone [4]
Dans le cas de l’acier 4340, nous avons une teneur en carbone (C) de 0.43 % environ.
Il s’agit donc d’un acier hypo-eutectoïde. La phase ferrite (α) possède une structure
cristallographique cubique centrée, alors que l’austénite (γ) possède une structure cubique
faces centrées. Pour réaliser le traitement thermique d’un acier, il est d’abord nécessaire de
le chauffer de façon à être complètement dans le domaine austénitique. Pour l’acier 4340,
une température de 1073 K suffit pour que l’acier soit complètement austénitique (voir
Figure 0.2).
4
0.2.1-Le traitement thermique et ses effets
Le traitement thermique consiste à refroidir rapidement un acier à l’état austénitique.
Au cours de ce refroidissement et en fonction de la vitesse de refroidissement, trois phases
peuvent apparaître : la perlite, la bainite et la martensite. Ces phases ont une influence sur la
dureté du matériau [5, 6]. La dureté d’un matériau se définit comme sa capacité à s’opposer
à la propagation d’une dislocation (défaut linéaire qui traduit une discontinuité dans
l’organisation de structure cristalline.). Le principe de traitement thermique de l’acier
repose sur la diffusion des atomes de carbone et le temps de diffusion accordé à ces atomes.
En effet, dans un acier austénitique, les atomes de carbone se logent dans les sites
interstitiels de la structure cubique à faces centrées de l’austénite. Lors du refroidissement,
ces atomes migrent vers les joints de grains s’ils en ont le temps. En dehors de la ferrite et
cémentite (Fe3C) pro-eutectoïde qui peuvent se former lors du refroidissement du matériau,
trois phases influencent principalement la dureté d’un acier : la perlite, la bainite et la
martensite.
0.2.1.1-La perlite
Il s’agit de la phase stable composée de lamelles de ferrite et de cémentite
superposées. Ces lamelles peuvent être observées sur la Figure 0.3. On obtient de la perlite
par un refroidissement lent de l’acier austénitique.
5
Figure 0.3 : Perlite observée au microscope 1000x [7]
La perlite peut être plus ou moins fine selon la vitesse de refroidissement. Plus le
refroidissement est rapide et moins les atomes de carbone ont le temps de se diffuser vers
les joints de grains et donc plus les lamelles de cémentite et ferrite sont fines. Une perlite
fine est plus dure qu’une perlite grossière. En effet, le nombre d’interfaces cémentite-ferrite
par unité de volume étant plus grand, les dislocations progressent plus difficilement, chaque
interface étant un obstacle à son avancée.
0.2.1.2-La bainite
Cette phase présente les mêmes constituants que la perlite (ferrite et cémentite), mais
possède une structure beaucoup plus fine sous forme d’aiguilles. La structure de la bainite
peut être observée sur la Figure 0.4. On obtient de la bainite lors d’un refroidissement trop
rapide pour former de la perlite (trop faible diffusion des atomes de carbone), mais pas
suffisamment pour obtenir de la martensite. La bainite est plus dure que la perlite.
6
Figure 0.4 : Bainite observée au microscope
0.2.1.3-La martensite
La martensite est une phase métastable de l’acier qui s’obtient par un refroidissement
très rapide de l’acier. Les atomes de carbone qui s’étaient logés dans les sites interstitiels de
la structure cubique à faces centrées de l’acier austénitique n’ont pas le temps de se
diffuser. Or, les sites interstitiels de la structure cubique centrée de la ferrite sont beaucoup
plus petits. Les atomes de carbones se retrouvent ainsi piégés dans des espaces trop petits
pour eux. Ils exercent alors des contraintes sur l’ensemble de la maille et la déforment. La
déformation de la maille engendre un grand nombre d’obstacles à la propagation d’une
dislocation. Cela explique la grande dureté observée dans le cas de la structure
martensitique. La microstructure de la martensite peut être observée sur la Figure 0.5.
1 µm
7
Figure 0.5 : Structure martensitique
0.3- LE TRAITEMENT THERMIQUE AU LASER
Ce traitement consiste à envoyer un rayonnement lumineux énergétique sur une zone
très localisée de la pièce. Il s’en suit d’un échauffement très localisé (et très rapide) de la
pièce sur une certaine profondeur. Une fois que le faisceau laser a quitté la zone frappée, le
volume réchauffé (souvent très petit) se refroidit très rapidement par diffusion de la chaleur
dans le reste du matériau. Dans le cas d’un traitement de surface d’un acier, le faisceau
laser transmet une importante quantité d’énergie dans un petit volume de matière sous la
surface. Ce volume de matière passe alors très rapidement au-dessus de la température
d’austénitisation de l’acier. Puis, ce volume, en se refroidissant très rapidement, engendre
l’apparition d’une structure martensitique particulièrement fine même pour des matériaux
difficiles à traiter thermiquement par des méthodes conventionnelles [2]. Le principe du
traitement laser est illustré sur la Figure 0.6. :
8
Vitesse d avance
Surface
Profondeur traitée
Faisceau laser
Figure 0.6 : Principe du traitement surfacique par laser
L’un des grands avantages de ce traitement est qu’il ne génère que très peu de
distorsions ou de contraintes internes [1, 2]. Deux critères influencent l’efficacité du
traitement de surface par laser. Pour les aciers, la zone à durcir doit être chauffée bien au-
dessus de la température d’austénitisation et maintenue suffisamment longtemps au-dessus
de celle-ci pour donner le temps nécessaire à la diffusion du carbone. De plus, les régions
en contact direct avec la région à durcir doivent être suffisamment importantes en volume
pour permettre le traitement thermique par diffusion de la chaleur vers ces régions.
Les paramètres clés du procédé sont [2] :
— La puissance de la source laser, la vitesse de balayage du laser et éventuellement les
caractéristiques du gaz protégeant la pièce de l’oxydation (argon ou hélium).
— Le modèle de distribution d’énergie (gaussienne, uniforme…).
— L’absorptivité du matériau pour la longueur d’onde du laser utilisé.
— Les propriétés thermiques du matériau (chaleur spécifique, conductivité thermique) et la
microstructure initiale.
9
Diminuer la vitesse de balayage conduit à une plus grande profondeur de traitement,
mais diminue la dureté de la surface. En effet, une plus faible vitesse de balayage signifie
une plus grande élévation de la température au niveau de la surface et donc un temps plus
important d’austénitisation. Ce temps plus important conduit à la formation de grains
d’austénite de grandes tailles et donc d’une microstructure plus grossière qui aura des
répercussions, après refroidissement, sur la microstructure de la martensite et sa dureté.
La Figure 0.7 montre, pour des vitesses de balayage différentes, un profil de dureté
pour un acier AISI 4340 revenu à 649 °C pendant 2 heures et chauffé avec un laser CO2
d’une puissance de 1.8 kW.
Figure 0.7 : Profils de dureté pour différentes vitesses de balayage dans le cas
d’un acier AISI 4340 [8]
Sur le schéma, on remarque, pour certaines vitesses de balayage, une zone ou la
dureté est inférieure à la dureté initiale du matériau. Cette zone correspond à une région non
traitée, mais affectée néanmoins par la diffusion de chaleur importante dans cette région
10
lors du refroidissement. Cela n’est en général pas désirable et il existe des moyens pour
limiter cet inconvénient.
Le mode de distribution d’énergie gaussienne (qui est admis comme étant le mode de
distribution naturelle de l’énergie laser) n’est pas toujours adapté lorsqu’il est question d’un
traitement de surface. En effet, la distribution non uniforme sur la tâche focale de l’énergie
peut causer la fonte localisée du matériau. Pour obtenir une distribution plus uniforme de
l’énergie sur une surface plus grande, on applique des méthodes de changement de forme
du rayon laser grâce à des lentilles qui permettent, notamment, de dé focaliser le rayon.
L’absorptivité du matériau joue également un rôle fondamental. Elle dépend
notamment de :
— la longueur d’onde de la source laser.
— L’angle d’incidence du rayon laser.
— Les propriétés thermiques du matériau et de sa microstructure.
— L’état de surface de la pièce (brute de fonderie, usinée, polie, meulée…) qui va jouer sur
le coefficient de réflexion de la surface.
Pour améliorer l’absorptivité d’un matériau, plusieurs solutions sont possibles. On
peut par exemple utiliser un très fin revêtement absorbant qui doit pouvoir s’adhérer à la
surface, ne pas agir chimiquement avec le matériau même chauffé et être retiré facilement
après le traitement.
0.4- AVANTAGES ET INCONVÉNIENTS DU PROCÉDÉ DE TRAITEMENT THERMIQUE
SURFACIQUE AU LASER.
Il existe de nombreuses façons de traiter thermiquement la surface d’un acier. Les
méthodes les plus courantes, autres que le laser, utilisent l’induction, le chalumeau, l’arc
électrique et le faisceau d’électrons. Le Tableau 0.1 ci-après montre les avantages et les
inconvénients de chacun des procédés [3]
11
Tableau 0.1 : Comparaison entre les différents procédés
Avantages Inconvénients
Laser
-Distorsions thermiques minimales
-Durcissement sélectif
-aucun refroidissant requis
-Possibilité de traiter les faibles
épaisseurs
-Profondeur de traitement contrôlable
-aucun post-traitement requis
-améliore la résistance en fatigue
-Équipements très onéreux
-Surfaces durcies limitées lors
d’une passe
-revêtements thermique souvent
requis
-de multiples passes provoquent
une fusion locale en surface.
Induction
-Production rapide
-Possibilité d’avoir de faibles
profondeurs traitées
-les coûts ont plus faibles que pour le
laser
-Surfaces de traitement importantes.
-Temps de pause obligatoire pour
changer les spires
-Distorsions thermiques
-faible flexibilité due à
l’importance de la place, de la
forme et de la taille des spires
-Fortes pénétrations thermales
-Dégradation possible de l’état de
surface
Chalumeau
-Peu onéreux
-flexible
-transportable
-Mauvaise reproductibilité
-Traitement thermique
relativement lent
-forte probabilité de distorsions
thermiques des pièces
-Problèmes environnementaux
Arc électrique Relativement peu onéreux et flexible
-Épaisseur de section limitée
-larges pénétrations thermales
Fusion de l’acier difficilement
contrôlable
12
Faisceau d’électrons
-Distorsions thermiques minimales -
Traitement sélectif
- Aucun refroidissant requis
-Équipements très onéreux
-faire le vide est requis
-faible productivité
-Coût important d’opération
0.5-PROBLÉMATIQUE GÉNÉRALE
Le procédé de traitement thermique au laser est de plus en plus employé dans
l’industrie automobile et aéronautique. Les composantes ainsi traitées telles que les
engrenages se distinguent par de meilleures performances comparativement à celles
produites par des procédés conventionnels [9]. Du point de vue industriel, le procédé
présente plusieurs avantages comparativement aux procédés de traitement
thermochimiques. Premièrement, le temps de traitement est très court et ne dépasse pas
quelques secondes. Deuxièmement, le traitement ne provoque pas de distorsions
thermiques majeures des composantes mécaniques développées du fait que la pièce est
superficiellement chauffée. Troisièmement, le procédé peut être facilement intégré dans des
cellules de fabrication. Finalement, le procédé ne fait usage d’aucun gaz et reste très
flexible comparé à la méthode de traitement par induction [1, 10].
Malgré les avantages du procédé, il reste difficile à intégrer à grande échelle dans les
chaînes de production. En effet, les compagnies manufacturières ont besoin d’atteindre de
grandes performances et d’avoir des recettes de développement fiables et peu couteuses
avant d’adopter un procédé. En effet, l’état actuel des connaissances et du savoir-faire
oblige les industriels à chercher les paramètres optimaux en se basant sur les méthodes par
essais et erreurs. Ces procédés industriels sont très couteux et demandent un grand temps de
développement. De plus, ils ne permettent pas d’obtenir une meilleure compréhension du
comportement global du procédé et de bien cerner tous les phénomènes en jeu [1]. En effet,
l’aspect multiphysique du procédé et les interactions entre le faisceau laser et le transfert de
chaleur affectent grandement le comportement du procédé [2]. De plus, plusieurs variables
telles que les paramètres du procédé et les propriétés du matériau affectent le comportement
13
global du procédé. Les propriétés du matériau varient en fonction de la température et de la
vitesse de chauffe. De plus, ces propriétés ne sont pas connues dans les conditions hors
équilibre thermodynamique. La complexité des interactions entre les phénomènes
thermiques, métallurgiques et mécaniques rend difficile l’établissement de modèles fiables
capables de prédire le profil de dureté en fonction des paramètres du procédé.
L’exploration systématique du procédé n’a jamais eu lieu en utilisant une approche
claire combinant l’étude des propriétés du matériau et des paramètres du procédé dans des
conditions hors équilibre thermodynamique [3]. Dans ce sens, la simulation par éléments
finis constitue un outil puissant pour mieux comprendre le comportement global du procédé
de traitement thermique au laser et pour quantifier l’effet des paramètres de simulation sur
la distribution de la température. Les modèles à développer doivent être basés sur les
méthodes numériques par éléments finis afin de résoudre les équations gouvernant le
champ thermique (Fourrier-Kirchhoff). La revue de littérature menée dans le cadre de ce
projet de recherche indique clairement que les recherches effectuées ont permis
d’approcher correctement le procédé de traitement thermique au laser, mais sans mettre en
place des approches capables de dégager concrètement les effets réels des propriétés du
matériau ou des paramètres du procédé sur le profil de dureté [4].
0.6-OBJECTIF
Le projet vise à étudier et surtout à prédire le profil de dureté en surface de pièces
mécaniques en acier, à géométries simples ou complexes, traitées thermiquement au laser.
Dans cette optique, il sera question de l’élaboration de modèles numériques 3D du procédé
en utilisant les éléments finis. Les résultats générés par ces modèles numériques seront
comparés aux essais expérimentaux.
14
0.7-MÉTHODOLOGIE
Dans un premier temps, il s’agit de pouvoir simuler avec un logiciel capable
d’effectuer un calcul par éléments finis le procédé de transformation surfacique au laser
dans le cas d’une géométrie simple. Le but de cette première étape est d’avoir une idée de
la nature des différents phénomènes physiques qui entrent en jeu dans ce procédé ainsi que
de leurs effets sur la microstructure de l’acier. Ensuite, le modèle doit aussi pouvoir générer
rapidement des résultats utilisables dans des études complémentaires afin d’avoir un
quelconque intérêt économique. Le modèle numérique doit bien sûr être robuste, c’est-à-
dire adaptable à n’importe quelle géométrie et avec n’importe quel ensemble de paramètres
de contrôle. Enfin, et surtout, les résultats donnés par le modèle numérique doivent être
vérifiés par des essais expérimentaux dans plusieurs cas de figure avec des paramètres de
contrôle différents.
Une fois un modèle numérique établi et vérifié dans le cas d’une géométrie simple, il
sera adapté aux géométries complexes que sont les engrenages. Les résultats dans le cas
d’une géométrie complexe devront aussi être vérifiés par des essais expérimentaux.
0.8-ORGANISATION DU MÉMOIRE
Le présent mémoire se divise en trois chapitres. Le premier chapitre se concentre sur
l’établissement d’un modèle de simulation en trois dimensions du procédé de traitement
thermique superficiel au laser dans le cas d’une géométrie simple. Un simple balayage de la
pièce dans le sens de sa longueur est alors réalisé avec une sélection spécifique de
paramètres (puissance, vitesse de balayage et diamètre du focus) différents à chaque
simulation. L’historique des températures est alors extrait du logiciel et utilisé pour
déterminer le profil de dureté grâce aux équations métallurgiques de Ashby et Easterling
[5]. Ensuite, ces profils de dureté simulés sont comparés aux profils de dureté réels
observés provenant de tirs au laser avec les mêmes ensembles de paramètres. De façon
générale, ce premier chapitre permet mieux comprendre les phénomènes thermiques qui
15
entrent en jeu dans le procédé de traitement thermique au laser et d’établir une
bibliographie sur l’ensemble des recherches effectuées sur le sujet.
Le deuxième chapitre reprend la géométrie simple traitée dans le premier chapitre,
mais utilise une autre modélisation : le laser est modélisé comme un flux de chaleur au lieu
d’une source de chaleur comme dans le premier chapitre. Cette modélisation à l’avantage
d’être plus facile à implémenter surtout en prévision d’un passage aux géométries
complexes. Le chapitre traite également de l’utilisation ultérieure des données de
simulation afin de réaliser des études statistiques précises ainsi que l’entrainement d’un
réseau de neurones. Le réseau de neurones, une fois entrainé grâce aux données de
simulation, est capable de donner les caractéristiques du traitement thermique selon les
paramètres d’entrée, et cela, indépendamment du logiciel de simulation.
Le troisième chapitre aborde les géométries complexes (plus particulièrement les
engrenages) et reprend la modélisation en flux de chaleur du deuxième chapitre. En outre,
dans ce chapitre, un mouvement de rotation de l’engrenage vient s’ajouter au mouvement
de translation du faisceau laser. Un banc d’essai incluant un tour en rotation a été
spécialement conçu et installé dans la cellule laser de l’UQAR pour le traitement thermique
des engrenages. Le procédé en lui-même contient une phase indispensable de préchauffage
puis une phase de traitement. Ce banc d’essai a permis la validation des résultats obtenus
lors de la simulation. Les résultats obtenus par le procédé sont assez comparables avec ceux
obtenus par le traitement par induction qui reste le procédé le plus utilisé aujourd’hui dans
l’industrie. Cependant, le procédé développé dans cette étude à l’avantage d’être robuste.
En effet, un seul montage permet le traitement d’engrenages de diamètres différents ce qui
n’est pas le cas avec le traitement par induction.
16
CHAPITRE 1
PRÉDICTION DU PROFIL DE DURETÉ D’UNE PLAQUE EN ACIER 4340
TRAITÉE THERMIQUEMENT AU LASER EN UTILISANT UN MODEL
3D ET UNE VALIDATION EXPÉRIMENTALE
1.1-RESUME EN FRANÇAIS DU PREMIER ARTICLE
L’article présente l’étude du traitement thermique au laser d’une plaque en acier 4340
au moyen d’un logiciel commercial de simulation. Le but de l’étude est d’être capable de
prédire quel sera le profil de dureté en fonction des paramètres de contrôle du procédé tels
que la puissance, la vitesse de balayage et le diamètre du faisceau) et cela, sans nécessiter
une étude expérimentale onéreuse. Ainsi, il sera possible d’obtenir la dureté en surface
désirée au premier essai et donc de faire des économies. La première partie de l’article se
concentre sur la mise en place du modèle 3D et de l’intégration dans le logiciel des
équations thermiques et métallurgiques. La deuxième partie concerne la génération de
différents profils de dureté avec différentes combinaisons de paramètres de contrôle. L’effet
de chacun des trois paramètres de contrôle sur le profil de dureté est alors analysé. Enfin,
dans la troisième partie, nous validons expérimentalement les résultats de simulation en
comparant, deux à deux, les profils de dureté réels et simulés pour différentes combinaisons
de paramètres. L’étude montre que les combinaisons de paramètres (puissance, vitesse de
balayage) doivent être choisies judicieusement pour permettre le traitement de l’acier sans
atteindre le point de fusion en surface. Globalement, les résultats obtenus sont satisfaisants
avec de faibles écarts entre profils simulés et profils réels.
17
Ce premier article, intitulé « Prediction of hardness profile of 4340 steel plate heat
treated by laser using 3D model and experimental validation », fut corédigé par moi-même
ainsi que par les professeurs Noureddine Barka et Abderrazak El Ouafi. Il fut accepté pour
publication dans sa version définitive en 2014 par l’American Society of Mechanical
Engineers (ASME). En tant que premier auteur, ma contribution à ce travail fut l’essentiel
de la recherche sur l’état de l’art, le développement de la méthode, l’exécution des tests de
performance et la rédaction de l’article. Les professeurs Noureddine Barka et Abderrazak
El Ouafi, respectivement second et troisième auteur, ont fourni l’idée originale. Ils ont aidé
à la recherche sur l’état de l’art, au développement de la méthode ainsi qu’à la révision de
l’article. Ahmed Chebak et Jean Brousseau ont contribué à la révision de l’article. Une
version abrégée de cet article a été présentée, sous forme d’affiche, à la conférence
International Mechanical Engineering Congress & Exposition (IMECE) à Montréal à
l’automne 2014.
18
1.2-PREDICTION OF HARDNESS PROFILE OF 4340 STEEL PLATE HEAT TREATED BY
LASER USING 3D MODEL AND EXPERIMENTAL VALIDATION
G. Billaud Mathematics, Computer and Engineering
Department
University of Quebec at Rimouski
Rimouski, Canada, G5L 3A1
N. Barka Mathematics, Computer and Engineering
Department
University of Quebec at Rimouski
Rimouski, Canada, G5L 3A1
A. El Ouafi Mathematics, Computer and Engineering
Department
University of Quebec at Rimouski
Rimouski, Canada, G5L 3A1
A. Chebak
Mathematics, Computer and Engineering
Department
University of Quebec at Rimouski
Rimouski, Canada, G5L 3A1
J. Brousseau
Mathematics, Computer and Engineering
Department
University of Quebec at Rimouski
Rimouski, Canada, G5L 3A1
1.2.1-abstract
This paper presents a study of hardness profile of 4340 steel plate heat treated by scanning
laser technique using 3D model. The proposed approach is carried out in three
distinguished steps. First, a 3D model is developed using an adequate formulation and
taking into account the nonlinear behaviour of the material. Then, the hardness curve is
approximated from the temperature distribution using metallurgical assumptions related to
the kinetic transformation and the temperature-time transformation diagram. Then, the case
depth is quantitatively analyzed versus the beam power density and scanning speed.
Finally, the developed approach is validated using experimental tests. The gap between
simulation and experimental results is determined. The obtained results allow predicting of
the hardness profile with a fairly good accuracy.
1.2.2-Nomenclature
𝐴𝑐1 Eutectoid temperature (K)
19
𝐴𝑐3 Austenitization temperature (K)
𝐴𝑐 Steel absorption coefficient (m-1)
𝑐 Carbon content of steel (wt %)
𝑐𝑐 Critical value of carbon composition
𝑐𝑒 Eutectoïde carbon composition (0.8% C)
𝑐𝑓 Carbon in ferrite (0.01% C)
𝐶𝑝 Specific heat (J/kg.K)
𝐷0 Pre-exponential of diffusion of carbon (m2/s)
𝑓 Volume fraction of martensite
𝑓𝑚 Maximum volume fraction of martensite
𝑓𝑖 Volume fraction occupied by the pearlite colonies
𝑔 Average grain size (µm)
𝐻𝑚 Hardness of the martensite (HV)
𝐾 Thermal conductivity (W/m.K)
𝐿𝑧 Plate thickness
𝑃 Laser power (W)
𝑄 Activation energy of diffusion of carbon (J/mol)
𝑅 Gas constant (J/mol.K)
𝑅𝑐 Reflection coefficient
𝑡 Time (s)
𝑇(𝑡) Heat cycle temperature (K)
𝑇0 Reference temperature (K)
𝑉0 Laser translational velocity (mm/s)
𝑤 Gaussian beam radius (mm)
𝜌 Density (kg/m3)
1.2.3-Introduction
Nowadays, laser surface hardening is becoming more and more used in aerospace
and automotive industries. This is due to the selected and local heat treatment character and
20
the high performances of mechanical component that it is capable of providing. Moreover,
the process provides hard surface layers with low levels of distortion in a very short time
[2, 3]. In fact, the distortions are less important than those obtained by flame, induction
heating and thermochemical hardening processes such as carbonizing and nitriding [11]. In
addition, the process offers the possibility to be fairly and precisely numerically controlled.
When the power density is applied and the scanning speed is adjusted, the laser beam
induces thermal flux within the part surface. Consequently, the initial microstructure is
transformed completely into austenite after reaching the austenitizing temperature (Ac3).
Then, it becomes martensite upon its self-quenching through heat conduction into the
colder bulk of the part core [12]. However, the heating stage must be long enough for the
carbon to be diffused and the austenitization to be completed. The quenching stage must be
fast enough to create fine martensite and to prevent the natural austenite degradation into
perlite or bainite [13]. Conversely, due to temperature gradients, it happens that below the
surface, as the depth increases, local and different austenitizing and quenching conditions
appears. This leads to dissimilar degrees of austenite homogenisation and thus, dissimilar
degrees of martensite homogenisation [14].
Despite the industrial advantages exhibited by the laser heat treatment process, it
still remains difficult to predict with a good accuracy the hardness profile based on the
process setting. Indeed, the process is affected by parameters such as input power, beam
size and scanning speed as well as the nonlinear properties of the material such as thermo-
physical and metallurgical properties [2, 3]. It is true that numerical simulation models
combining heat conduction modes and metallurgical transformations can be advantageous
to understand the qualitative and quantitative influences of each parameter on the process.
However, they are not accurate enough and reality representative to predict results such as
hardness profiles, case depths and compressive residual stresses with a good accuracy.
Since the heat treatment process consists of a fast laser heating, it is difficult to
characterize all physical quantities such that the generated heat amount and the temperature
distribution during surface transformation. The process is so fast that the measurement
devices such as the thermocouples or pyrometers are not able to follow the process
21
dynamic due to the time constant. The interference effects affect the quality of measured
signal playback. In addition, the calibration of this instrument is performed under specific
conditions and depends greatly on the material emissivity.
Literature review conducted during this work allows synthetizing the development
efforts done in the laser heat treatment field. Many studies propose some numerical models
based on thermal and metallurgical equations in order to establish a relationship between
the case depth and the process parameters variation [15]. The hardness profile is closely
depending on the temperature distribution and the metallurgical transformations occurred
during heating. These models could be exploited in order to predict hardness profile under
some specific conditions. The thermal model usually resolves the transient heat based on
heat-flow equation to find the temperature distribution [2, 16]. Generally, the Finite
Element Method (FEM) is used to solve the equation with time dependant parameters [15-
17]. There is an interesting exception where a Finite Difference Method (FDM) is used to
solve the thermal equation which is less time consuming but also less accurate [18]. Once
the temperature distribution is determined, metallurgical equations allow predicting the
phase transformation and thus the hardness profile. In this way, the Johnson-Mehl-Avrami
and Koistinen-Marburger equations are used to predict the phase transformation which are
based on the discretization of the Time-Temperature-Transformation (TTT) diagram [16].
Moreover, equations from Ashby and Easterling are used as the metallurgical model to
determine the phase transformation [19, 20]. Once the phases’ proportions are computed,
the Maynier equations are usually used to calculate the final hardness profile using a
mixture rule [21]. Note that the result of those simulations usually do not show the over
tempered zone, where the hardness becomes inferior to the initial hardness of material,
which can be highlighted by the experimental study under certain initial microstructural
properties [8]. Finally, there is no based-simulation study for the laser surface hardening
process of the AISI 4340 steel and the studies usually do not compare the effectiveness of
the process with other hardening process like induction, nitriding, etc.
This paper aims to predict the hardness profile of plate made in 4340 steel and
hardened by laser process as a function of the beam power and the scanning speed. The
22
proposed 3D model takes into account the machine parameters and the material properties
[2]. The developed model is validated using experimental tests under various operation
conditions. In this work, the commercial COMSOL® software is used to solve the heat
transfer equation using finite element method (FEM). The developed model is then
combined to the MATLAB® software to determine the hardness profile using the
metallurgical equations. The non-linear behaviour of the material is considered in that
model. The FEM analysis allows taking into account the temperature dependence of certain
properties like the thermal conductivity and the specific heat. The simulation is then
compared and validated by experimental results. The gaps between the simulation and
experimental results are statistically determined.
1.2.4-Formulation
1.2.4.1-Thermal conduction
Based on heat conduction mode, the developed models depend on the laser type and
usually used the governing Fourier-Kirchhoff heat flow equation [2, 15-17]. The
temperature distribution recorded at a specific time and known spatial coordinates is
obtained by solving this well-known equation. Once the temperature distribution is
available, the hardness profile can be determined by using metallurgical transformation
equations. Ashby and Easterling established mathematical equations for metallurgical
transformations during laser surface transformation of hypoeutectoid steel [5]. The most
commonly used formula is the transient heat conduction equation in a solid. The heat
transfer can be described by the Fourier-Kirchhoff equation.
( ) ( , , , )T
Cp K T E x y z tt
(1)
𝞺, Cp and K, denote, respectively, the density, specific heat and thermal
conductivity of the material. x, y, z denote the Cartesian coordinates. The term E (x, y, z, t)
represents the heat source created by the laser beam. In this study, the heat source used for
the simulation is a Gaussian beam distribution which is given by equation 2 [22].
23
2 20
2 2
( ( 0 )) (y y0)( )
( )2 22
(1 ) e2
x x V t
Ac Lz zw wAc
E P Rc ew
(2)
w is the Gaussian beam radius, P is the laser beam power, Rc is the reflection
coefficient of the surface of the material treated by laser [3], Ac is the beam power
absorption coefficient [2]. V0 is the scanning speed, Lz is the part thickness and x0, and y0
are the beam center coordinates at t = 0 s. E (W/m3) represents a Gaussian heat source
moving according to the x-axis at the speed V0.
1.2.4.2-Metallurgical transformation
As the temperature in the material reaches the eutectoid temperature Ac1, the perlite
colonies transform almost instantaneously into austenite while the proeutectoid ferrite
remains, assuming the initial microstructure is composed of perlite and ferrite. If it is
tempered martensite, then the martensite transforms into austenite. This is the perlite
dissolution into austenite (Figure 1.1). After the perlite colonies are transformed into
austenite, the carbon diffuses outward from these transformed zones into the proeutectoid
ferrite which increases the volume fraction of austenite (Figure 1.2). This process is called
the homogenisation of austenite.
24
Figure 1.3 : Transformation from perlite into
austenite [5]
Figure 1.4 : Homogenization of hypoeutectoid steel [5]
The total number of diffusive jumps which occurs during the heat cycle affects the
extent of the structural change and is given by the kinetic strength [5].
I= exp -Q/ R×T t dt (3)
Q is the activation energy for the transformation and R is the gas constant. It is more
convenient to express I as described in equation 4.
25
pI=ατ×exp -Q/ R×T (4)
Tp is the peak temperature at the considered depth and τ is the thermal time constant. The
term α and τ are approximated by the equations 5 and 6.
pα=3 R×T /Q (5)
1
p 0τ= 1-Rc P/ 2πKe V T -T (6)
T0 is the initial temperature.
Ashby and Easterling established that if the perlite plate spacing is designed as “l”
then for a lateral diffusion of the carbon over the distance “l” would be enough to complete
the transformation from perlite into austenite and that for a heat cycle when temperature is
time dependent [5].
2
02 exp( )p
Ql D
R T
(7)
D0 is the diffusion constant for the carbon in ferrite.
The obtained austenite has the same carbon content as the perlite ce = 0.8%. From
there, the carbon diffuses in the proeutectoid ferrite. When the temperature reaches Ac1, the
volume fraction of austenite is the volume fraction fi previously occupied by the perlite
colonies (and the minimum volume fraction of martensite).
i f ff = C-C / 0.8-C C/0.8 (8)
cf is the negligible carbon content of the ferrite and c is the carbon content of the steel.
Thus, the maximum martensite fraction allowed by the transformation temperature
time diagram (TTT) is
26
p 1
i p 1 3 1 1 p 3
fm = 0 if T < Ac
fm = fi+ 1-f T -Ac / Ac -Ac if
®®®®®®
Ac T Ac
p 3 ®®®®®® fm = 1 if T > Ac
(9)
Ac3 is the temperature of complete austenitization (Table 1.1).
Ashby and Easterling supposed that all the material with a specific carbon
proportion more than a critical value Cc will transform into martensite. The volume fraction
of the martensite is then given by equation 10 [2, 5].
2/3
i i e c 0f=fm- fm-f ×exp - 12f / g π ×ln C / 2C D ×I (10)
g is the mean grain size.
The hardness can then be calculated by a mixture rule.
m f+pH=f×H + 1-f ×H (11)
The value Hm and Hf+p are given by the Maynier equations that take in account the
cooling rate and the composition [21].
1.2.5-Simulation
The laser heat treatment applied to an AISI 4340 50 mm x 40 mm x 5 mm
parallelepiped plate is simulated with published thermal data measured at thermodynamic
equilibrium. The laser has a Gaussian energy distribution in the spot created on the plate
surface. The simulation efforts are performed using a 3D software. The simulation
parameters considered for the modeling are the power (in W) and the scanning speed (in
mm/s). The material properties behavior versus temperature is pondered in this study since
finite element analysis is able to integrate the real behavior in the computation process.
These physical properties are only known at thermodynamic equilibrium and transient
phenomena taking place during fast heating or cooling are not taken into consideration. The
material is considered as homogeneous and isotropic. The FEM analysis takes into account,
27
as boundary limits the ambient temperature set at 293 K. Table 1.1 shows the material
properties used for the simulation aside from the specific heat and thermal conductivity.
Table 1.1 : Material properties
Property Symbol Unit Value
Reflection coefficient Rc 0.6
Steel absorptivity Ac m-1 800
Eutectoid temperature Ac1 K 996
Austenitization
temperature Ac3 K 1053
Austenite grain size
(assumed) g m 10
Activation energy of
carbon diffusion in ferrite Q kJ/mol 80
Pre-exponential for
diffusion of carbon D0 m2/s 6x10-5
Gas constant R J/mol.K 8.314
Steel carbon content C 0.43 %
Austenite carbon content Ce 0.8 %
Ferrite carbon content Cf 0.01 %
Critical value of carbon
content Cc 0.05 %
Volume fraction of pearlite
colonies fi 0.5375
Based on equation 2, the model is implemented using Heat transfer module of
COMSOL® with x0 = 0 mm and y0 = 15 mm. The mesh is built to be fine upon the laser
path and coarse elsewhere (Figure 1.3). A convergence study and has allowed using a
suitable mesh density at the region where the laser beam heats the surface. Figure 1.3
shows the final mesh configuration used in this study.
28
Figure 1.5 : Final used mesh (dimensions are in mm)
Figure 1.4 shows the effect of mesh quality on the obtained temperature. The
temperature TA, TB and TC are chosen to have a representation of the temperature
distribution at the surface. The three points have the same coordinate according to the x-
axis (25 mm) and the same coordinate according to the z-axis (5 mm). According to the y-
axis, the three measurement locations A, B and C are at 12.5 mm, 15 mm and 17.5 mm
respectively. The temperature TB is at the surface directly on the middle plan where the
laser beam is in contact with the surface. The temperatures TA and TC are positioned
symmetrically across the middle plan. Due to the symmetry appearance, these temperatures
should display the same value. The convergence curves demonstrate that the temperatures
become unchanged between the size mesh of 0.6 mm and 1.1 mm. In the outside of this
range, the temperatures fluctuate due to the accuracy and truncation errors.
29
Figure 1.6 : Effect of mesh size on obtained temperatures
The final mesh size that is chosen is 1.1 mm upon the laser path and is a good
compromise between accuracy and computation time. The input power, scanning speed and
reflection coefficient of the surface can be freely arranged. The simulation takes into
account the emissivity of the surfaces as well as the heat convection phenomenon through
the surfaces to the ambient air. The obtained results are developed at specific process
parameters for the beginning in order to understand the effect and to present the
temperature distribution.
For this purpose, the power is fixed at 1200 W and the scanning speed at 15 mm/s.
The simulation results show that the temperature reach high values in the region near to the
spot path and decreases drastically away from it. To illustrate the temperature behavior, 3
cases are exposed. Figure 1.5, Figure 1.6 and Figure 1.7 show the isothermal contours at
1 s, 2 s and 3 s respectively. At the beginning (Figure 1.5), at the beam spot center, high
value of temperature are recorded (above 1100 C) and the generated heat remains
concentrated around the spot. At the end of the heating cycle (3 s), high values of
temperature are still recorded at the beam spot center. However, the part becomes globally
heated with more heat amount due to the conduction. Some heat is lost by convection and
radiation modes through the surface but the major part created by the laser is absorbed by
200
400
600
800
1000
1200
1400
0.1 0.6 1.1 1.6 2.1 2.6 3.1 3.6 4.1 4.6
Tem
per
atu
re(
C)
Mesh size (mm)
TA
TB
Tc
30
the part by conduction. The fact that the laser concentrates the power in a small area
generates a very high heat flux with high temperature gradient. It affects the metallurgical
transformation especially in the region around the spot beam. Once heated, this region
could reach melting point and the hardness profile will include a small non desirable
melting layer. However, the part volume presents an industrial advantage since it acts as a
cooler to transform the region heated above the Ac3 directly to hard martensite without a
forced convection.
Figure 1.7. Isothermal contours of temperature (t = 1 s)
31
Figure 1.8. Isothermal contours of temperature (t = 2 s)
Figure 1.9. Isothermal contours of temperature (t = 3 s)
Figure 1.8 shows the temperature evolution depending on the time at different
depths under the surface. The temperature curves are compared to Ac1 and Ac3
temperatures to approximate the case depth. The temperature rapidly increases to reach a
maximum value of 1050 C before dropping off. The temperature profile shows that the
32
temperature is important at the surface and decreases with the depth. At only 3 mm depth,
the temperature at the same time (1 s) is three times lower. The figure also shows that the
cooling rate is extremely high which is essential for the martensite to be formed. In the case
of AISI 4340, the time taken to drop at a low temperature is well within the limit of the
TTT diagram which ensures the formation of the hard martensite microstructure due to its
high hardenability. A first approximation could confirm that the case depth is between 0.7
mm and 1.4 mm for the chosen process parameters.
Figure 1.10 : Temperature versus time for various depths
measured from heated surface at middle plan
Figure 1.9 shows the Gaussian temperature profile along the y-axis at a given time
and at different depths and the comparison with the temperatures Ac1 and Ac3. It can be
seen that away from the middle axis (beam spot center), the region heated beyond the
austenitization temperature Ac3 becomes smaller. Under the midpoint, the case depth is
about 1.5 mm but it is only 0.5 mm if a 1.3 mm shifted point to the left or to the right is
considered. The assumption stipulates that all austenitized regions become martensite upon
cooling. In fact, the used steel is recognized to have a good hardenability [23]. The case
depth is about 1 mm in this case.
200
400
600
800
1000
1200
1400
0 1 2 3
Tem
per
ature
(°C
)
Time (s)
0 mm 0.7 mm 1.4 mm 2.1 mm
2.8 mm Ac1 Ac3
33
Figure 1.11 : Temperature profile along the y-axis at a given
time and at different depths
1.2.6-Validation
To validate the developed model, a 3 kW laser power is used (IPG YLS-3000-ST2).
The laser is equipped with a 6 degrees of freedom articulated robot. The experimental setup
is illustrated in Figure 1.10. The plate is placed according a referential coordinates
systems. In order to be at the maximum power focus, the upper surface of the plate is
adjusted vertically to match the focal distance. The laser beam diameter is evaluated at 1.08
mm when focused. The focal distance is 310 mm with permissible tolerance of ± 5%. The
programmed tests consist of heating the surface with chosen scanning speed and power to
generate a martensitic transformation within a specific case depth more than 0.5 mm. At
each distinct test, the laser beam travels a distance of 40 mm along the plate length. Once
the treatment is carried out and the samples are prepared and polished, a micro-hardness
measurement allows characterizing the hardness curve as a function of the depth using a
Clemex® micro-hardness machine.
Prior to the laser heat treatment, the samples are oil-quenched then tempered in a
furnace at 510 C during 1 hour. As a result, the initial microstructure is a tempered
200
400
600
800
1000
1200
1400
10 11 12 13 14 15 16 17 18 19 20
Tem
per
ature
(°C
)
Width (mm)
0 mm 0.5 mm 1 mm
1.5 mm Ac1 Ac3
34
martensite with hardness about 400 HV. In the case of AISI 4340, the tempering treatment
before the laser heat treatment has been proved to have a huge impact on the final hardness
profile. For steels tempered at a high temperature, the over tempered zone, where the
hardness of the steel is inferior to the core hardness of the material (and which is generally
undesirable), will be minimized [8].
Figure 1.12 : Experimental setup
The validation test is based on a factorial design composed of 2 factors at 2 levels.
The first factor is the laser beam power and it is varying from 950 W to 1100 W. The
second factor is the scanning speed varying from 16 mm/s to 18 mm/s. Table 1.2 shows the
experimental matrix (L4) for laser hardening validation.
Table 1.2 : Experimental matrix for validation
Test Power (W) Scanning speed (mm/s)
1 950 16
2 950 18
3 1100 16
4 1100 18
Figure 1.11 shows a transverse section of the laser-hardened zone. As expected of a
Gaussian distribution where the energy is higher at the center of the laser beam spot, an
35
elliptic section can be observed. The case depth is approximately 0.72 mm and the width of
this region is 2.34 mm. The hardened region can be divided in melting and hard region.
Figure 1.13 : Metallographic picture of hardness profile -
Test 2
Figure 1.12 to Figure 1.15 show the predicted and measured hardness curves. To
describe the hardness evolution versus the depth, the core hardness 𝐻𝑓+𝑝 is replaced by the
material core hardness (400 HV). It is important to note that even with an initial tempered
martensitic microstructure instead of an initial perlite microstructure, the measured
hardness is always in average within a 15% uncertainty of the predicted value as shown in
Table 1.3. The case depth hardened is between 0.7 mm for the set of parameters 950 W and
18 mm/s. The case depth is evaluated at 1.05 mm for a power of 1100 W and a scanning
speed of 16 mm/s. As expected, the case depth seems to increase when the laser power
increases or the scanning speed is reduced. It is relevant to remark that, each time, just
before the transition where the hardness drops; there is a small area where the hardness is
higher than it is just below the surface. In fact, the surface hardness is not maximal since
36
this region is melted. The hardness continues to increase with the depth to reach maximal
value or hard martensite before dropping off.
After this area, there is another area called the over-tempered zone which has not
been treated but has been nonetheless affected by the heat and softened. In this area, the
hardness is less than it is in the core material. Usually unwanted, the size of this zone
depends on the initial microstructural properties [8]. However, the model used in this study
is unable to predict the two precedent areas that are observed for all the tests conducted in
this study.
Figure 1.14 : Hardness curve for test 1 (950 W and 16 mm/s)
200
300
400
500
600
700
800
0 200 400 600 800 1000 1200 1400 1600
Har
dnes
s(H
V)
Depth (mm)
Measured hardness Predicted Hardness
37
Figure 1.15 : Hardness curve for test 2 (950 W and 18 mm/s)
Figure 1.16 : Hardness curve for test 3 (1100 W and 16 mm/s)
200
300
400
500
600
700
800
0 200 400 600 800 1000 1200 1400 1600
Har
dnes
s (H
V)
Depth (mm)
Measured hardness Predicted hardness
200
300
400
500
600
700
800
0 200 400 600 800 1000 1200 1400 1600
Har
dnes
s (H
V)
Depth (µm)
Measured hardness Predicted hardness
38
Figure 1.17 : Hardness curve for test 4 (1100 W and 18 mm/s)
In Figure 1.14 and Figure 1.15, It can be seen that the measured hardness under the
surface is less than the predicted hardness (a difference of 40 HV approximately). This gap
can be explained by the fact that the temperature just under the surface most likely reached
the melting point during the experiments with a power of 1100 W due to the fact that the
reflection coefficient is a bit lower than predicted. For each experiment, the maximum
relative error seems to be high but this only is because the model is unable to predict the
hardness in the over tempered zone which results in a located big gap between the
simulated hardness and the measured hardness in this zone. However, as the mean relative
error testifies, for each test, the rest of the predicted hardness profile fairly matches the
measured profile (Table 1.3).
Table 1.3 : Average absolute and relative hardness errors
Absolute error (HV) Relative error (%)
Test 1 25 4.9
Test 2 28 5.1
Test 3 40 8.4
Test 4 58 9.2
200
300
400
500
600
700
800
0 200 400 600 800 1000 1200 1400 1600
Har
dnes
s (H
V)
Depth (µm)
Measured hardness Predicted Hardness
39
Table 1.4 shows the comparison between hardened depths for each experiment.
Because the reflection coefficient is difficult to determine with a good accuracy and it has
an impact on the case depth, some small differences can be observed between measured
case depth and simulated case depth. As the drop of hardness is very fast in the transition
zone, even an error of 0.1 mm between simulated case depth and measured case depth leads
to a big gap between simulated hardness and measured hardness in this zone which can be
especially seen on Figure 1.15. The absolute difference between the measured depth and
the predicted depth is less than 0.1 mm for each case. The model can predict the case depth
with maximal relative error of 14.3 %.
Table 1.4 : Comparison of hardened depth
Measured case
depth (mm)
Predicted case depth
(mm) Relative error (%)
Test 1 0.8 0.7 -12,5
Test 2 0.7 0.6 -14,3
Test 3 0.9 1.0 -11,1
Test 4 0.9 0.8 -11,1
1.2.7-Conclusion
In this study, a laser heat treatment applied to AISI 4340 steel is modeled using a
FEM commercial software. The model generates successfully numerical solutions of the
heat-flow equation and gives a complete simulated 3D temperature distribution in the laser-
hardened sample. It allows an accurate prediction of the hardness profile without doing any
experiments. As seen in the validation paragraph, the generated model fairly matches both
the actual hardness and the depth. A maximum hardness of 734 HV with a case depth of
1.05 mm is obtained in these experiments. However, both the simulation and the
experiment show that the input power and the scanning speed must be carefully chosen in
order to harden a part on a satisfying depth without reaching the melting point at the
surface of the material. The obtained results will be advantageously exploited to determine
the sensitivity of the developed model using statistical tools within large range of variation
of the model.
40
CHAPITRE 2
MODÈLE DE RÉSEAU DE NEURONES ARTICFICIELS POUR
L’ESTIMATION DU TRAITEMENT THERMIQUE SUPERFICIEL AU LASER
D’UNE PLAQUE D’ACIER 4340
2.1-RESUME EN FRANÇAIS DU DEUXIEME ARTICLE
L’article se concentre sur l’amélioration du modèle développé dans le premier
article afin de le rendre plus robuste en vue du passage plus aisé à des géométries
complexes. Il s’agit d’utiliser l’équation de la chaleur afin de déterminer la distribution
thermique dans le matériau pendant le traitement puis de déterminer le profil de dureté
grâce aux équations d’Ashby et Easterling. La différence avec le premier article est qu’ici,
la source de chaleur est remplacée par un flux de chaleur plus proche de la réalité d’un
faisceau laser. Cette nouvelle approche est vérifiée par la suite avec des tests
expérimentaux. De plus, l’article aborde une utilisation ultérieure des données (profondeurs
et largeurs durcies) générées par le modèle. En effet, une fois le modèle numérique mis en
place et vérifié expérimentalement, il permet d'effectuer un grand nombre de simulations
avec des paramètres d'entrée différents. Ce grand nombre de données permet de réaliser des
études statistiques très précises et fiables (notamment sur les effets relatifs des paramètres
sur le résultat du traitement thermique superficiel au laser). En outre, ces résultats
permettent d’entrainer un réseau de neurones artificiels qui peut alors fonctionner
indépendamment d’un logiciel de calcul par éléments finis et donner instantanément le
résultat d’un traitement thermique superficiel au laser pour un ensemble de paramètres
donné.
Ce deuxième article, intitulé «ANN Based Model for Estimation of Transformation
Hardening of AISI 4340 Steel Plate Heat-Treated by Laser», fut rédigé par moi-même ainsi
42
que par les professeurs Noureddine Barka et Abderrazak El Ouafi. Il fut accepté pour
publication dans sa version finale en 2015 par le « Journal of Materials Sciences and
Applications (MSA) ». En tant que premier auteur, ma contribution à ce travail fut
l’essentiel de la recherche sur l’état de l’art, le développement de la méthode, l’exécution
des tests de performance et la rédaction de l’article. Les professeurs Abderrazak El Ouafi et
Noureddine Barka, respectivement second et troisième auteur, ont fourni l’idée originale.
Ils ont aidé à la recherche sur l’état de l’art, au développement de la méthode ainsi qu’à la
révision de l’article.
43
2.2- ANN BASED MODEL FOR ESTIMATION OF TRANSFORMATION HARDENING OF
AISI 4340 STEEL PLATE HEAT-TREATED BY LASER
G. Billaud Mathematics, Computer Science and
Engineering Department
University of Quebec at Rimouski
Rimouski, Canada, G5L 3A1
A. El Ouafi Mathematics, Computer Science and
Engineering Department
University of Quebec at Rimouski
Rimouski, Canada, G5L 3A1
N. Barka Mathematics, Computer Science and
Engineering Department
University of Quebec at Rimouski
Rimouski, Canada, G5L 3A1
2.2.1-Abstract
Quality assessment and prediction becomes one of the most critical requirements for
improving reliability, efficiency and safety of laser surface transformation hardening
process (LSTHP). Accurate and efficient model to perform non-destructive quality
estimation is an essential part of the assessment. This paper presents a structured and
comprehensive approach developed to design an effective artificial neural network (ANN)
based model for quality estimation and prediction in LSTHP using a commercial 3 kW
Nd:Yag laser. The proposed approach examines laser hardening parameters and conditions
known to have an influence on performance characteristics of hardened surface such as
hardened bead width (HBW) and hardened depth (HD) and builds a quality prediction
model step by step. The modeling procedure begins by examining, through a structured
experimental investigations and exhaustive 3D finite element method simulation efforts, the
relationships between laser hardening parameters and characteristics of hardened surface
and their sensitivity to the process conditions. Using these results and various statistical
tools, different quality prediction models are developed and evaluated. The results
demonstrate that the ANN based assessment and prediction proposed approach can
effectively lead to a consistent model able to accurately and reliably provide an appropriate
44
prediction of hardened surface characteristics under variable hardening parameters and
conditions.
Keywords: Laser hardening process, AISI 4340 steel, case depth, hardened
bead width, artificial neural network.
2.2.2-Introduction
In the industry, many steel components require a surface heat treatment in order to
have the desired surface qualities such as hardness and wear resistance. Among the
available processes, laser hardening process (LHP) is one of the most efficient, as it allows
a very fast and localized metallurgical transformation. In addition, the process generates a
hard surface layer with low distortion [2]. Using high energy beam, the surface is rapidly
heated to reach the transition temperatures (microstructure changes) before being quenched
by heat conduction into the colder core of the material. Consequently, a martensitic layer is
produced without affecting the core of the material [3].
Despite its industrial advantages, predicting hardness profiles with a good accuracy
remains difficult. Indeed, besides the process parameters (Power, scanning velocity and
focus diameter), which can be properly set, the process is affected by the non-linear
behavior of thermo-physical and metallurgical properties of the material [2]. It makes the
temperature distribution uneasy to predict by complicating the resolution of the governing
heat-flow equation. Experimental tests are also expensive in terms of time and resources,
especially if one wants to test many combinations of control parameters to have a better
understanding of the process.
Among all the approaches that can be used to understand the process and ultimately
to predict its performance, 3D simulation represents a powerful tool for combining multi-
physics problems and taking into account the material and complex geometries. In fact, the
developed model includes the non-linear properties of the material and the heat-flow
equation is solved using the finite element method (FEM) [14, 17]. The FEM enables
45
solving the governing heat-flow equation that determines the temperature distribution for
each time step during the heating process. The hardness is then approximated by the
equations of Ashby and Easterling [5]. The advantage of the simulation is that, although it
might be long and tedious to implement, once it is completed and experimentally verified in
a few cases, one can test any combination of input parameters and quickly generate a large
number of data which can be used for further exploitation. Many studies have been
conducted to optimize the various laser process parameters (surface hardening, laser
welding, laser cutting, etc.) through statistical methods such as the ANOVA method. It can
be applied in many fields of engineering, including production processes and products for
professional and consumer markets all over the world [19]. S-L Chen and D. Shen [24]
used the Taguchi tools such as graphic designs of parameters and analysis of variance
(ANOVA) to optimize the hardened depth (HD) and the hardened bead width (HBW) in the
case of the LHP. Badkar and Pandey [20] used the same tools to determine the relative
importance of each parameter on the LHP. K.Y. Bentounis, A.G. Olabi and M.S.J. Hashmi
[25] conducted a similar study in the case of a laser welding process. Most recently, Sathiya
et al. [26] also used the Taguchi method to optimize the laser welding parameters. Given
that experimental characterization requires great efforts in terms of time and money, it is
not easy to experiment all the combinations of the input parameters. The Taguchi method
really is an asset, as it is a partial factorial design that only requires some combinations of
the input parameters in order to be performed and yet, it gives accurate statistical results in
the overall process.
Others studies are conducted using artificial neural networks (ANN) in order to
improve the performance of the laser processes [27]. Ciurana J. et al. [28] used ANNs to
establish a model for laser micromachining of hardened steel and to optimize the process
parameters. Pan Q.Y et al. [29] performed a similar study by using a neural network to
model the non-linear relationship between laser processing parameters and corrosion
resistance of the surface of stainless steel during the process of laser surface re-melting,
which locally improved the corrosion resistance of the steel. Munteanu and Adriana [30]
predicted the surface hardness of steel using a neural network in the case of an electron
46
beam machining process which is similar to the LHP. F. Lambiase et al. developed a
prediction model of laser hardening by means of an ANN using experimental datasets and
linear interpolations between those experimental measures to train the network. However,
to obtain good and efficient modeling results with ANN techniques, a large quantity of
experimental data is advantageous and the observations should cover a sampling space as
wide as possible in order to simplify the interpolation task.
Indeed, in any modelling experiment, the results depend, to a large degree, on the
method used to collect data. In a lot of cases, full factorial experiments are conducted. This
approach cannot be implemented when too many factors are under consideration, because
the number of repetitions required would be prohibitive in time and cost. Regular fractional
factorial designs cannot produce credible results when interactions among the factors exist.
By contrast, the use of a testing strategy such as the orthogonal arrays (OAs) developed by
Taguchi leads to an efficient and robust fractional factorial design of experiments that can
collect all the statistically significant data with a minimum number of repetitions.
Accordingly, OAs are used in this study for the experimental design. On the other hand, by
using 3D FEM simulation that can provide results matching fairly well with the
experimentally observed variables, one can easily and quickly obtain additional data for
any combination of input parameters. The quantity of simulated data generated in a short
time compared to experimental data would allow exhaustive statistical analysis including
all levels of all input parameters. Moreover, with a large quantity of data for training, a
simple Multilayer Perceptron ANN can be appropriate for modeling.
The objective of this paper is to presents a structured and comprehensive approach
developed to design an effective artificial neural network (ANN) based model for quality
estimation and prediction in LSTHP using a commercial laser source. The proposed
approach examines laser hardening parameters and conditions known to have an influence
on performance characteristics of hardened surface such as hardened bead width (HBW)
and hardened depth (HD) and builds a quality prediction model step by step. The modeling
procedure begins by examining, through a structured experimental investigations and
47
exhaustive 3D FEM simulation efforts, the relationships between laser hardening
parameters and characteristics of hardened surface and their sensitivity to the process
conditions. Using these results and various statistical tools, different quality prediction
models are developed and evaluated. In order to carry out the models building procedure,
an efficient modeling planning method combining neural networks paradigm, a multi-
criteria optimization and various statistical tools is adopted.
2.2.3-3D Model implementation and validation
2.2.3.1-Implementation
The 3D FEM model is developed on the commercial software to estimate the
temperature profiles. These temperature profiles are used to approximate the surface
hardness profiles (surface hardness, HD, HBW). The part is a 50 mm × 30 mm × 5 mm
parallelepiped (Figure 2.1).
Figure 2.1 : Sample with its mesh implemented on COMSOL
In this study, the heat flux used for the simulation is considered as a Gaussian beam
distribution type which is given by equation 1 [2],
48
2 22 2
0 0 0E=E ×exp{-( x-(x +V×t) / 2w + y-y / 2w )} (1)
Where V is the scanning velocity, and x0 and y0 are the beam center coordinates at t
= 0 s. E (W/m2) represents a Gaussian heat flux moving according to the x-axis at the
velocity V. E0 is defined by equation 2,
2
0E =P 1-Rc / 2πw (2)
Where w is the Gaussian beam radius, P is the laser beam power, Rc is the reflection
coefficient of the material surface [3].
The moving isothermal contours can be observed in Figure 2.2. Because of the
Gaussian form of the beam, the temperature is at its maximum (about 1110 K) at the center
of the spot. The temperature decreases rapidly with the depth because of heat conduction
into the colder core of the material. The heated volume is small at the beginning of the
process (t = 0.4 s) and it gets larger as the time passes and the beam moves (t = 2.5 s). As it
can be seen in Figure 2.2, the small volume of the part that had reached the temperature of
1110 K at t = 0.4 s (see Figure 2.2.a) cool downed to reach a temperature under 430 K at t
= 2.5 s (see Figure 2.2 b). It means that a very fast quenching happened in that volume.
Once the temperature distribution is determined, the hardness profile is estimated
using the equations of Ashby and Easterling [5]. Those equations are implemented in
MATLAB® to obtain the hardness at any point belonging to the heated part and,
consequently, the hardness curve representing the hardness versus depth.
The 4340 steel properties are displayed in Table 2.1.
The specific heat and the thermal conductivity are temperature dependant and their
dependency is taken into account in our model.
49
(a)
(b)
Figure 2.2 : Isothermal contours: (a) --» t = 0.4 s and (b) --» t = 2.5 s
Table 2.1 : Metallurgical properties
Property Symbol Unit Value
Reflection coefficient Rc 0.6
Eutectoid temperature Ac1 K 996
Austenitization temperature Ac3 K 1053
Austenite grain size (assumed) g m 10
Activation energy of carbon diffusion in ferrite Q kJ/mol 80
Pre-exponential for diffusion of carbon D0 m2/s 6x10-5
Gas constant R J/mol.K 8.314
Steel carbon content C 0.43 %
Austenite carbon content Ce 0.8 %
Ferrite carbon content Cf 0.01 %
Critical value of carbon content Cc 0.05 %
Volume fraction of pearlite colonies fi 0.5375
2.2.3.2-Metallurgical equations
When the temperature in the material reaches the eutectoid temperature Ac1 in a small
volume under the surface, the steel microstructure, which is generally tempered martensite
in the case of the steel AISI 4340, starts to transform into austenite. The complete
transformation from tempered martensite to austenite occurs when the temperature reaches
Ac3. In the case of laser hardening treatment, when the temperature drastically decreases
50
due to rapid heat diffusion into the colder core of the part, the austenite transforms into hard
martensite. This is what is called a heat cycle (Figure 3).
As seen on Figure 3, the peak temperature at the surface is above Ac3. Therefore, a
complete transformation into hard martensite happened at the surface. However, the peak
temperature at 1.4 mm under the surface is under Ac1. It means that no transformation
happened at this depth.
The total number of diffusive jumps that occur during the heat cycle affects the extent
of the structural change and is given by the kinetic strength I [2, 5],
I = exp -Q/ R×T t dt (3)
Where Q is the activation energy for the transformation and R is the gas constant. It is
more convenient to express I as described in equation 4.
pI = α τ×exp -Q/ R×T (4)
Here Tp is the peak temperature at the considered depth and τ is the thermal time
constant. The terms α and τ are approximated by equation 5 and 6,
pα = 3 R×T /Q (5)
1
p 0τ = 1-Rc P/ 2 π K e V T -T (6)
Where T0 is the initial temperature.
51
Figure 2.3 : Temperature evolution at different depths (850 W and 9 mm/s)
The obtained austenite has the same carbon content as a perlite microstructure Ce =
0.8 %. From there, the carbon diffuses into the proeutectoid ferrite. When the temperature
reaches Ac1, the volume fraction of austenite is fi (which is also the minimum volume
fraction of martensite), given by equation 7,
i f ff = C-C / 0.8-C C/0.8 (7)
Where Cf is the negligible carbon content of the ferrite and C is the carbon content of
the steel.
The maximum martensite fraction allowed by the transformation temperature time
diagram (TTT diagram) is
p 1
i p 1 3 1 1 p 3
fm = 0 if T < Ac
fm = fi+ 1-f T -Ac / Ac -Ac if
®®®®®®
Ac T Ac
p 3 ®®®®®® fm = 1 if T > Ac (8)
200
400
600
800
1000
1200
1400
0 1 2 3
Tem
per
ature
(°
C)
Time (s)
0 mm0.7 mm1.4 mm2.1 mm2.8 mmAc1Ac3
52
Ashby and Easterling supposed that all the material with a specific carbon proportion
above the critical value Cc will transform into martensite. The volume fraction of the
martensite is then given by equation 9 [2, 5].
2/3
i i e c 0f = fm- fm-f ×exp - 12 f / g π ×ln C / 2 C D ×I (9)
Here g is the mean grain size and D0 is the diffusion constant for the carbon in ferrite.
The hardness can then be calculated by a mixture rule (equation 10).
m f+pH=f×H + 1-f ×H (10)
The value Hm and Hf+p are given by Maynier equations that take in account the
cooling rate and the composition of the material [21].
2.2.4-Experimental validation
The experimental procedure consists of a first heat treatment in a furnace with a water
quenching followed by a tempering at 640 °C for 1 hour. The aim is to reach a
homogeneous hardness of 440 HV for all the samples. Then, a commercial 3 kW Nd:Yag
laser power (IPG YLS-3000-ST2), combined with a 6 degrees of freedom articulated robot
(Figure 2.4) is used to perform laser heating. The plan-parallel sample is put on a metal
plate under the laser head. This type of laser generates a laser beam with a wavelength λ =
1064 µm. The process parameters are the input power, the scanning velocity and the focus
diameter. In this study, the laser beam has a straight-line trajectory as seen in Figure 2.2.
Finally, the resulting case depth is measured by micro indentation.
53
Figure 2.4 : Experimental setup for model validation
Experimental validations are conducted according the Table 2.2. The focus diameter
is 1260 µm for the three tests. The values are chosen so that the surface temperature reaches
the austenite temperature Ac3 but does not hit the melting temperature (about 1450 °C).
Table 2.2 : Experimental matrix for validation
Test Power (W) Scanning velocity (mm/s)
1 850 9
2 850 12
3 950 12
A micro-hardness machine is used to characterize the hardness curve as a function of
the depth. After the laser treatment, the samples are prepared and polished to reach
adequate surface finish. The hardness is then measured by using a micro-hardness machine.
The validation is conducted by micro indentation, with 100 µm steps between consecutive
Vickers marks on the surface along a vertical axis. The experimental results help to validate
and calibrate the model. In this sense, the obtained results confirm the concordance
between the experimental and simulated hardness curves. This suggests that even if the
developed model is not able to accurately predict the hardness curve, it can determine the
hardened depth with good accuracy. Figures 2.5, Figure 2.6 and Figure 2.7 show a
54
comparison between the simulated and measured hardness curve using Vickers hardness
scale (VH) for the three tests (Table 2.2). It is worth noting that the developed 3D model is
unable to predict the over-tempered zone where the hardness of the material becomes
inferior to the initial hardness. However, hardened zone, transition zone and unaffected
zone are correctly predicted. As expected, the hardened depth (at the start of the transition
zone) increases as the power rises and/or the scanning velocity decreases. Table 2.3 shows
the average absolute and relative errors between measured and simulated hardness. The
preliminary tests allow to conclude that, despite the difference of more than 50 HV in terms
of absolute error, the relative error is very small, not exceeding 10 %. As shown in Figure
2.5, Figure 2.6 and Figure 2.7, the simulation is fairly accurate in both hardness prediction
and case depth prediction.
Table 2.3 : Average absolute and relative hardness errors resulting from the preliminary
tests
Test Absolute error (HV) Relative error (%)
1 43 8.8
2 30 5.2
3 24 4.0
Figure 2.5 : Hardness curve for test 1 (850 W and 9 mm/s)
300
400
500
600
700
800
0 200 400 600 800 1000 1200 1400
Har
dnes
s (H
V)
Depth (µm)
Simulated hardness (HV)
Measured hardness (HV)
55
Figure 2.6 : Hardness curve for test 2 (850 W and 12 mm/s)
Figure 2.7 : Hardness curve for test 3 (950 W and 12 mm/s)
300
400
500
600
700
800
0 200 400 600 800 1000 1200 1400
Har
dnes
s (H
V)
Depth (µm)
Simulated hardness (HV)
Measured hardness (HV)
300
400
500
600
700
800
0 200 400 600 800 1000 1200 1400
Har
dnes
s (H
V)
Depth (µm)
Simulated hardness (HV)
Measured harness (HV)
56
2.2.5-Calibration of the model with corrected Rc
The coefficient Rc can be estimated around 0.6 for steel [3]. However, this coefficient
greatly varies according to the surface temperature. Moreover, the surface temperature
depends on the process parameters. In order to correctly calibrate Rc, different
combinations of process parameters are executed using laser heating cell and the Rc is
corrected so the results generated by the simulation match the experimental results for each
set of input parameters. Finally, Rc is approximated as a function of the process parameters
using a linear regression technique. Table 2.4 shows the Rc values depending on the laser
power, the scanning velocity and the focus radius of the beam spot. The coefficient seems
to increase as the power and/or the scanning velocity increases. Also, it seems to decrease
when the focus radius increases.
Table 2.4 : Corrected Rc according to process parameters
Power P (W) Scanning velocity V (mm/s) Focus radius Rad (µm) Corrected Rc
400 20 550 0.50
520 20 550 0.54
630 20 550 0.57
740 20 550 0.61
400 12 550 0.47
400 16 550 0.49
400 16 480 0.51
400 16 613 0.48
400 16 663 0.47
The regression equation (equation 11) proves that there is a linear relationship
between Rc and the process parameters. The correlation coefficient is 0.994, which
confirms a good correlation.
Rc=0.4205+0.000303×P+0.003553×V-0.000198×Rad (11)
P is the input power in W, V is the scanning velocity in mm/s, and Rad is the focus
radius in µm. The developed equation is incorporated in the simulation model. The HD and
HBW can be estimated with good accuracy as a function of the process parameters.
57
2.2.6-Shadowgraph measurements
As this study is focused on the HD and HBW and not the hardness values themselves,
the depth and width are measured using optical method based on shadowgraph
measurement. Figure 2.8 shows a micrographic picture of a part heat treated by laser with a
power of 1000 W and a scanning velocity of 12 mm/s. The hardened region with hard
martensite appears very clearly after a chemical treatment and can even be observed with
the naked eye. Two significant zones can be distinguished. The first one is the melted
region near the surface that received a great amount of energy, enough to reach the melting
point. The second region represents the hardened region where the temperature exceeded
the austenitization temperature (Ac3) without reaching the melting point and where the
microstructure changed into martensite upon self-quenching.
Figure 2.8 : Micrographic picture illustrating the HBW and
HD after chemical etching
2.2.7-Statistical study
In the present study, the objective is to predict the HD and HBW with given process
parameters provided by the great number of data generated through simulation (assuming
Hardened bead width (HBW)
Hardened depth (HD)
1 mm
58
the input parameters are included in the range of study). A statistical study is conducted
through a design of experiment (DOE) to determine the relative significance of each
parameter and the interactions between them. The ANOVA method aims to study the
effects of parameters on the hardness. It gives the contribution of each parameter on the
variation of the outputs (HD and HBW). The process parameters and their design levels are
displayed in Table 2.5. The levels are chosen so that the surface transformation happens
and the surface temperature does not hit the melting temperature regardless of the
combination of process parameters.
Table 2.5 : Factors and levels used for the ANOVA study
Factors Factor Levels
Laser Power (P) [W] 410 520 613 740
Scanning Velocity (V) [mm/s] 12 16 18 20
Focus Radius (Rad) [µm] 480 550 613 663
The simulation allowed us to quickly obtain results for all 64 (43) possible
combinations of factor levels, and thus to generate a full factorial design. Statistical studies
such as analysis of variance, main effects studies and linear regression are conducted.
2.2.7.1-Anova for HD versus P, V and Rad
Table 2.6 presents the detailed statistical analysis. An F-value above 11.77 implies
that the parameter is very significant. In this case, power (P), scanning velocity (V), focus
radius (Rad) are all significant models terms. The interaction terms are less important since
their contributions are less than 0.4 %. Also, it is clear that the power and the scanning
velocity have the largest effect on the response value and that they are equivalent with
contributions around 48 %. The three interaction terms can be considered negligible.
59
Table 2.6 : ANOVA for HD
Source DF SS contribution MS F-value p-value
P 3 563906 48.4 % 187969 499.71 0.000
V 3 568906 48.8 % 189635 504.14 0.000
Rad 3 13281 1.1 % 4427 11.77 0.000
P×V 9 4531 0.4 % 503 1.34 0.264
P×Rad 9 2656 0.2 % 295 0.78 0.632
V×Rad 9 2656 0.2 % 295 0.78 0.632
Model 36 1155936 99.1 % 383124
Error 27 10158 0.9 % 376
Total 63 1166094
Figure 2.9 shows the effect of all parameters on the case depth (HD). The obtained
results confirm that the HD increases as beam power increases and/or as scanning velocity
decreases. It also increases as the focus radius decreases. The ANOVA method is
conducted in order to assess the significance of each parameter. For each parameter studied,
the variance ratio value, F, is compared to the values from standard F-tables for given
statistical levels of significance. In this way, it is concluded that within the investigated
processing ranges, the power, the scanning velocity and the focus radius are significant for
the case depth at 95 % confidence. Since the interaction terms have negligible
contributions, they will not be considered in the rest of the study. Figure 2.10 shows the
HD calculated using the regression formula (equation 12) for all 64 combinations of
process parameters and their distribution around the bisector of the quadrant. If the formula
is perfectly accurate, all the points should be on the bisector. For the regression formula to
be considered accurate, a maximum relative error of 10 % is allowed for all 64 sets of
process parameters. A maximum relative error of 6.51 % is observed, with a mean relative
error of 2.25 % between the HD calculated with the regression formula and the one
simulated by the software. The coefficient of determination R2 is mainly used to measure
the relationship between experimental data and measured data. A coefficient R2 = 99.13 %
indicates an accurate study.
HD=1113.7+0.7557×P-31.70×V-0.2036×Rad (12)
60
Figure 2.9 : Effects of parameters on case depth
2.2.7.2-Anova for HBW versus P, V and Rad
Table 2.7 shows the detailed statistical analysis. An F-value above 70.68 implies that
the parameter is very significant. In this case, power (P), scanning velocity (V), focus
radius (Rad) are all significant models terms. The interaction terms are less important since
their contributions are less than 0.7 %. Also, it appears that the input power and the
scanning velocity have the largest effect on the response value with contributions around
37-43 %. The three interaction terms can be considered negligible. The coefficient of
determination R2 is mainly used to measure the relationship between experimental data and
measured data. Just like for the hardened depth, the input laser power and the scanning
velocity have the same degree of impact (and the opposite effect); the other parameter
(Rad) still have significance, and the interactions are negligible.
700
800
900
1000
1100
410 520 630 740
Mea
nP
12 14 16 18 20
V
480 550 613 663
Rad
61
Figure 2.10 : Comparison between simulated HD and HD calculated by
regression formula (equation 12)
Table 2.7 : Results of the ANOVA for HBW
Source DF SS Contribution MS F-value p-value
P 3 1763125 37.4 % 587708 161.71 0.000
V 3 2023125 43 % 674375 185.56 0.000
Rad 3 770625 16.4 % 256875 70.68 0.000
P×V 9 13125 0.3 % 1458 0.40 0.923
P×Rad 9 30625 0.7 % 3403 0.94 0.511
V×Rad 9 10625 0.2 % 1181 0.94 0.959
Model 36 4611250 98 % 1525000
Error 27 98125 2 % 3634
Total 63 4709375
600
700
800
900
1000
1100
1200
1300
600 700 800 900 1000 1100 1200
Cal
cula
ted
HD
(µ
m)
Simulated hardened depth (µm)
Statistically calculated HD (µm)
Simulated HD (µm)
62
Figure 2.1 : Main effects plot for hardened bead width
The first thing one can notice on Figure 2.11 is that the main effects plot for HBW is
similar to the main effects plot for the HD, with the noticeable exception of the focus
radius, which has the opposite effect on the HBW compared to the effect it has on the HD.
Indeed, when the focus radius increases the HBW increases as well, whereas the HD
decreases (see Figures 2.9 and Figure 2.11). This is caused by the Gaussian distribution of
the energy at the surface of the material, which results in a relationship between HD and
HBW. Indeed, the fact that the radius is greater while the power and the scanning velocity
remain the same means that there will be less energy at the center of the focus.
As for the HBW, it appears that the interactions are negligible with very low F-value.
Therefore they will not be included in the regression equation.
Figure 2.12 shows the HBW calculated using the regression formula (equation 13)
for all 64 combinations of process parameters and their distribution around the bisector of
the quadrant. If the formula is perfectly accurate, all the points should be on the bisector. A
maximum relative error of 6.95 % is observed, with a mean relative error of 2.39 %
between the HBW calculated with the regression formula and the one simulated by the
software. Both values are well under the maximum criteria of 10 % and thus, the formula
can be considered accurate.
2200
2300
2400
2500
2600
2700
2800
410 520 630 740
Mea
nP
12 14 16 18 20
V
480 550 613 663
Rad
63
Moreover, the coefficient of determination R2 = 97.92 % testifies an accurate
regression equation albeit not as satisfying as it is for the HD.
HBW=1782+1.3409×P-57.68×V+1.598×Rad (13)
Figure 2.12 : Comparison between simulated HBW and HBW calculated by
regression formula (equation 13)
In addition to the statistical study, and in order to provide a reliable alternative to
standard thermal techniques that would be accurate and less time consuming, we conducted
a study with an artificial neural network (ANN).
2.2.8- Neural network modeling
As compared to other techniques, an ANN provides a more effective modeling
capability, particularly when the relationship between sensor-derived information and the
characteristic(s) to be identified is non-linear. ANNs can handle strong non-linearity, a
large number of variables, and missing information. Based on their intrinsic learning
capabilities, ANNs can be used in a case where there is no exact knowledge concerning the
1900
2100
2300
2500
2700
2900
3100
3300
1900 2100 2300 2500 2700 2900 3100 3300
Ca
lcu
late
d H
BW
(µ
m)
Simulated hardened bead width (µm)
Statistically calculated HBW (µm)
largeur durcie en surface (µm)
64
nature of the relationships between various variables. This is very useful in reducing
experiment efforts.
A neural network is used to predict the hardened depth and hardened bead width.
Neural networks are generally presented as systems of interconnected neurons, where the
links between neurons are weighted. Figure 2.13 shows the general principle of an ANN
model. The goal is to produce one or more outputs that reflect the user-defined information
stored in the connections during training.
In this study, a Generalized Feed-Forward Multilayer Perceptron (GFF-MLP) neural
network model with one hidden layer containing 7 neurons is chosen. While various ANN
techniques can be used in this approach, generalized feed forward networks seem to be the
most appropriate because of their simplicity and flexibility. Before selecting the variables
and training the models, it is important to establish the network topology and optimize the
training performances. The idea is to approximate the relationship between the network
parameters and the complexity of the variables to be estimated. The selected network is that
which achieved the best results, the [n | 2n+1 | 3] network, where n is the number of inputs.
The perceptron is characterized by a nonlinear sigmoid function. This type of neural
network is always fully connected, meaning each perceptron of each layer is connected
with all the perceptrons in the previous layer [31]. In a GFF-MLP network, connections
between layers can jump over one or more layers. In practice, these networks solve
problems much more efficiently than MLP networks [32].
Neural networks need to be trained with data sets in order to be able to interpolate for
any given input parameters that fall within the training range. Neural networks cannot
extrapolate, which means one cannot get reliable outputs if the input parameters are not
within the range of the training parameters. In this study, the goal is to obtain a neural
network able to predict the case depth and hardened bead width for a given combination of
input parameters (within its training range). In all neural networks, during the training step,
65
the input data are normalized to the range of [-1, 1]. The weights and biases of the network
are initialized to small random values to avoid a fast saturation of the activation function.
ANN model
Laser Power (P)
Scanning Speed (V)
Focus Radius (rad)
Hardened Depth (HD)
Hardened Bead Width (HBW)
Figure 2.13 : Principle of the neural network
2.2.8.1-Maintaining the integrity of the specifications
For a commercial laser device, there are usually 3 control parameters, the input power
(P), the scanning velocity (V) and the focus radius (Rad). In this study, 4 levels for each of
those parameters are chosen and are displayed in Table 2.5. The levels are chosen to ensure
minimal martensitic transformation and to avoid the melting point (about 1450°C)
regardless of the combination of levels. With 3 parameters with 4 levels, the total number
of possible combinations is 64 (43). The simulation allows to quickly get all of the 64
combinations and produce a full L64 matrix as in the preceding statistical studies.
In addition to the training data, a neural network also requires verification data (that
are different from the training data) in order to validate the training step. These verification
data are displayed in Table 2.8. In this case, the mean value of two consecutive levels are
identified and used in simulation to generate data for verification. This leads to a validation
design of 33 possible.
The neural network is trained considering the mean square error (MSE) of the cross-
validation as an achievement indicator. The training of the neural network stops when the
MSE stops decreasing. In order to evaluate the effectiveness of the network, some criteria
are used, the correlation coefficient and the root mean square error, which would be
respectively equal to 1 and 0 in the best case scenario with perfect accuracy.
66
Table 2.8 : Middle points
Factors Factor Levels
Laser Power (P) [W] 465 575 685
Scanning Velocity (V) [mm/s] 14 17 19
Focus Radius (Rad) [µm] 515 581.5 638
2.2.8.2-Result and interpretation
Once the training step of the network is performed, the 27 combinations of
verification data are applied as input parameters. The outputs of the ANN model are
compared with those obtained by simulation. Therefore, this comparison is effective using
various statistical indexes that characterize the prediction capability of the ANN model.
Two main criteria are used to evaluate the accuracy of the network: the absolute error and
the relative error.
Figure 2.14 shows the absolute errors for both HD and HBW for all 27 test
combinations. The maximum absolute errors for HD and HBW are, respectively, 64 and 94
µm. This means that the absolute error is of less than 100 µm for the overall test data, for
both HD and HBW. Given that the values of HD are between 700 µm and 1100 µm, and
that the values of HBW are between 2400 µm and 3000 µm, the model exhibits a good
potential in terms of accuracy.
67
Figure 2.14 : Absolute relative errors for HD and HBW
As can be seen in the Figure 2.15, the relative errors for both the HD and HBW are
very low in every case. The maximum relative errors for HD and HBW are, respectively,
8.01 % and 3.62 %. The mean relative errors for HD and HBW are 2.40 % and 1.63 %,
respectively, which heightens the accuracy of the neural network.
Figure 2.15 : Relative errors for HD and HBW
0
10
20
30
40
50
60
70
80
90
100
1 3 5 7 9 11 13 15 17 19 21 23 25 27
Ab
solu
te e
rro
rs (
µm
)
Trial
absolute errors HD
absolute errors HBW
0%
1%
2%
3%
4%
5%
6%
7%
8%
9%
1 3 5 7 9 11 13 15 17 19 21 23 25 27
Rel
ativ
e er
ror
Trial
Relative errors HD
Relative errors HBW
68
Figures 2.16 and Figure 2.17 present, respectively, the results of the ANN models
during the verification stage for HD and HBW. In fact, the figures show the ANN model
and those obtained by simulation. The data are mostly located around the bisector of the 1st
quadrant, which outlines the accuracy of the model. The two figures show that the network
is well trained and is highly efficient. The network is therefore a reliable way to predict the
HD and HBW for any combination of input parameters within the training range (between
480 W and 663 W for power, 12 mm/s and 20 mm/s for scanning velocity, 480 mm and
663 mm for focus radius). The ANN models don’t require any computation time to predict
the outputs comparatively to the simulation. Note that the ANN models can predict the
desired outputs in the studied variation range only and they cannot extrapolate outside.
Figure 2.16 : Comparison between simulated HD and HD
calculated by the neural network.
650
750
850
950
1050
1150
700 800 900 1000 1100
Pre
dic
ted
HD
by n
eura
l net
wo
rk (
µm
)
Simulated HD (µm)
69
Figure 2.17 : Comparison between simulated HBW and HBW
calculated by the neural network
Table 2.9 shows the comparison between the results generated by simulation and
those generated by ANN model during training stage and confirms the observations from
Figure 2.14 and Figure 2.15.
Table 2.9 : Comparison of the results
Simulation ANN models Absolute error Relative error
HD
(µm)
HBW
(µm)
HD
(µm)
HBW
(µm)
HD
(µm)
HBW
(µm)
HD
(%)
HBW
(%)
Training
Min 700 2000 698 2058 2 58 0.03 0.05
Max 1200 3200 1175 3147 25 53 4.88 4.52
Mean 937 2557 931 2570 15 39 1.62 1.55
Verificat
ion
Min 700 2200 730 2164 1 2 0.10 0.09
Max 1050 3000 1084 2964 64 94 8.01 3.62
Mean 894 2522 902 2521 21 40 2.40 1.63
Even if the ANN models have good performances in terms of robustness and
accuracy, it is still important to validate them using experimental validation.
2000
2200
2400
2600
2800
3000
3200
2200 2400 2600 2800 3000
Pre
dic
ted
HB
W b
y n
eura
l net
wo
rk (
µm
)
Simlated HBW (µm)
70
2.2.9-Experimental validation of the neural network
The great number of data that can be generated by a 3D finite element simulation
allows to accurately train a neural network that will be able to predict the HD and HBW,
and thus, it avoids the need to produce expensive experimental data that are often less
numerous because of their cost.
Once the network accuracy is verified with data generated by a FEM simulation,
experimental validation tests are conducted using a Nd:Yag laser and the shadowgraph
measurement method.
Eight sets are randomly chosen among the 27 sets of verification data. The
experimental matrix is displayed in Table 2.10.
The input powers are between 465 W and 685 W, the scanning velocities are between
14 mm/s and 17 mm/s. Finally, the focus radii are between 515 µm and 638 µm.
Table 2.10 : Experimental matrix for validation
Test Power (W) Scanning velocity (mm/s) focus radius (µm)
1 465 14 515
2 575 19 515
3 685 17 515
4 685 14 515
5 575 14 581.5
6 685 17 581.5
7 465 17 638
8 685 17 638
The results of the tests are shown in Table 2.11. The maximum relative errors for
both the HD and HBW are 7.37 % and 2.93 %, respectively.
The ANN is able to correctly predict both HD and HDW. It can now be used
independently from the COMSOL software. It is easier to use as one only needs to compute
71
the process parameters (within the training ranges of the ANN) to obtain reliable results
instantly.
Table 2.11 : Experimental validation – results
Test Network
HD (µm)
Experimental
HD (µm)
Relative
error for
HD (%)
Network
HBW
(µm)
Experimental
HBW (µm)
Relative
error for
HBW (%)
1 929 909 2.2 2444 2375 2.88
2 864 894 3.35 2302 2320 0.76
3 968 953 1.59 2470 2402 2.83
4 1084 1010 7.37 2726 2653 2.74
5 997 1018 2.04 2735 2702 1.22
6 952 1022 6.86 2603 2591 0.46
7 777 785 0.98 2392 2324 2.93
8 936 1002 6.55 2715 2643 2.73
2.2.10-Conclusion
In this paper, a structured and comprehensive approach developed to design an
effective ANN-based model for quality assessment and prediction in LSTHP using a
commercial 3 kW Nd:Yag laser is presented. Several laser hardening parameters and
conditions were analyzed and their correlation with multiple performance characteristics of
hardened surface was investigated using a structured experimental investigations and
exhaustive 3D FEM simulations under consistent practical process conditions. Following
the identification of the hardening parameters and conditions that provide the best
information about the LSTHP operation, tow type of modeling techniques were proposed to
assess and predict the hardened bead width and hardened depth (HD) of the laser
transformation hardened AISI 4340 steel plate: multiple regression analysis and ANN
approach. The results demonstrate that the regression approach can be used to achieve a
relatively accurate predicting model with correlation larger than 90 %. The ANN models
present greater results. The maximum relative errors for both HD and HBW are less than 8
% and 3 %, respectively. Globally, the performance of the ANN-based model for quality
estimation and prediction in LSTHP shows significant improvement as compared to
conventional methods. With a global maximum relative error less than 10 % under various
72
LSTHP conditions, the modeling procedure can be considered efficient and have led to
conclusive results, due to the complexity of the analyzed process. The proposed approach
can be effectively and gainfully applied to quality assessment for LSTHP, because it
includes the advantages of ease of application, reduced modeling time and sufficient
modeling accuracy.
CHAPITRE 3
ANALYSE THERMIQUE DU TRAITEMENT SURFACIQUE AU LASER DES
ENGRENAGES-SIMULATION 3D ET VALIDATION
3.1-RESUME EN FRANÇAIS DU TROISIEME ARTICLE
L’article aborde le cas de la simulation numérique du procédé de traitement
thermique superficiel au laser dans le cas d’une géométrie complexe (ici des dents
d’engrenage). Les résultats, qui sont la profondeur durcie au sommet et au pied des dents,
donnés par la simulation sont ensuite vérifiés par des essais expérimentaux. L’engrenage
est monté sur un tour en rotation et le faisceau laser balaye ensuite l’engrenage sur toute sa
largeur. Le modèle numérique simule ce traitement avec un flux de chaleur à travers la
surface du matériau auquel on impose un mouvement de rotation et de translation. Il est
vite apparu qu’un préchauffage avec de rapides balayages laser en va-et-vient à faible
puissance est nécessaire pour amener la température des dents proche de la température
d’austénitisation. Un unique balayage final à forte puissance permet alors de traiter les
dents en profondeur. Les profils de traitement obtenus par la simulation et la validation
expérimentale sont très similaires à ceux obtenus par le procédé de traitement par induction
couramment utilisé dans l’industrie de nos jours pour le traitement des dents d’engrenages.
Ce troisième article, intitulé « Thermal analysis of surface transformation
hardening of gears using laser – 3D simulation and validation », fut corédigé par moi-
même ainsi que par les professeurs Abderrazak El Ouafi et Noureddine Barka,
respectivement second et troisième auteur. En tant que premier auteur, ma contribution à ce
travail fut l’essentiel de la recherche sur l’état de l’art, le développement de la méthode,
l’exécution des tests de performance et la rédaction de l’article. Les professeurs Abderrazak
74
El Ouafi et Noureddine Barka ont fourni l’idée originale. Ils ont aidé à la recherche sur
l’état de l’art, au développement de la méthode ainsi qu’à la révision de l’article.
75
3.2- THERMAL ANALYSIS OF SURFACE TRANSFORMATION HARDENING OF GEARS
USING LASER – 3D SIMULATION AND VALIDATION
G. Billaud Mathematics, Computer Science and
Engineering Department
University of Quebec at Rimouski
Rimouski, Canada, G5L 3A1
A. El Ouafi Mathematics, Computer Science and
Engineering Department
University of Quebec at Rimouski
Rimouski, Canada, G5L 3A1
N. Barka Mathematics, Computer and
Engineering Department
University of Quebec at Rimouski
Rimouski, Canada, G5L 3A1
3.2.1-Abstract
This paper presents the study of the laser surface transformation hardening of a low
alloy AISI 4340 steel workpiece of finite width such as a gear. A 3D model implemented
on a commercial software enables solving the heat transfer equation, the determination of
the temperature distribution under the surface as well as the prediction of optimal process
parameters. The laser beam is considered as a moving disc heat flux through the surface,
with a Gaussian distribution of heat intensity. With a temperature distribution that can be
determined at any given time during the process, the width and depth of the hardened zone
beneath the surface can be determined using the Ashby and Easterling metallurgical
equations.The process parameters are the laser power, P, the laser beam focus
diameter, 𝐷𝑓𝑜𝑐 , and the traverse scanning velocity, 𝑉𝑠𝑐. In this study the gears are mounted
on a lathe, and the laser beam is immobile and placed above the rotating gear. This adds
another control parameter, rotation speed, 𝑊𝑟. A commercial Nd:Yag laser beam is used to
validate the simulated results. The results demonstrate that the numerical simulation can
effectively lead to a consistent model able to accurately and reliably provide an appropriate
prediction of hardened surface characteristics of gear teeth under variable hardening
parameters, without requiring an actual expansive trial and error procedure. In addition, the
76
results of the experimental procedure conducted in this study appears to be very
comparable with those obtained by the induction process.
Keywords: Laser transformation hardening process, Nd:YAG laser, Hardened depth,
Hardened bead width, gears.
3.2.2 -Introduction
The type of microstructure under the surface that forms during manufacturing affects
the fatigue performance of mechanical components. The wear resistance of steel
components, such as gears or bearings, depends on their surface hardness [33]. Different
types of surface hardening processes are used to achieve the required hardness, such as
induction hardening or case carburization. Another method is to use a laser beam. Lasers
are becoming more and more widely used in industrial applications including cutting,
drilling, welding and heat treatment (among many others) [2]. The laser hardening process,
which is the subject of the present study, can be fully computer-controlled or numerically
controlled. In the case of steel components with complex geometries such as gears, a laser
beam allows a very localized and selective hardening with control of treated depth and
width [2]. Moreover, the localized heating reduces the thermal distortions to a minimum.
The rapid heating allows the transformation of the steel into austenite before it is
transformed into hard martensite upon cooling, due to rapid heat conduction into the colder
core of the steel (self-quenching). Surface melting is to be avoided in laser surface
hardening of steels. With the melting temperature of low alloy steels like AISI 4340 being
around 1430 °C, it limits either the maximum power density (which depends on the laser
power and the beam focus diameter) or the minimum scanning velocity. Nevertheless, the
power density must be high enough and the interaction time must be long enough for the
steel to transform from its initial microstructure into austenite. This sets either the
minimum power density or the maximum scanning velocity. For AISI 4340 the transition
temperature from the initial microstructure to complete austenite is at about 800 °C. In the
case of laser surface transformation, the cooling rate is generally above 103 °C/s, which is
more than enough for the martensite to be produced in most low alloy steels, including
77
AISI 4340 steel. The main process parameters are generally the laser power, scanning
velocity and beam diameter. Optimizing and reducing the laser heat treatment cost is an
issue in aerospace and automotive industries. Because experimentation is expensive in the
case of laser heating, it is better to be able to predict the results of laser heat treatment
depending on the parameters.
3.2.3-Literature review
Steen and Courtney [34] conducted a study of the laser transformation of AISI 1036
steel using a 2 kW continuous wave CO2 laser. A five level, full factorial design
experiment was conducted with power levels from 1.2 kW to 2 kW, scanning velocities
from 25 mm/s to 400 mm/s, and beam radii from 1.6 mm to 5.8 mm. With these experiment
designs, they statistically established a relationship between the hardened depths and the
process parameters.
Ashby and Easterling [5] conducted a study of the transformation hardening of hypo-
eutectoid steels using a laser beam. They used 500 W and 2500 W continuous CO2 lasers
with both Gaussian and “top hat” energy profiles. They combined approximate solutions
for the heat flow with kinetic models to predict the change of microstructure and hardness
with depth. Finally, they produced diagrams which showed the beam power density, beam
radius, scanning velocity, depth below the surface for a given microstructure, and the
resulting hardness profiles.
Following the work of Ashby and Easterling, Davis et al. [13] included kinetic
models describing the change of microstructure in their thermal analysis of a rectangular
workpiece.
Steen and Mazumder [3] established a three dimensional heat transfer model for laser
material processing, with a moving Gaussian distribution heat source, by using finite
difference numerical techniques. In their study, they considered the system to be in quasi-
steady-state conditions, in that the thermal profile was considered to be steady relative to
the position of the laser beam.
78
Generally, the thermal properties of carbon steels, such as thermal conductivity and
specific heat, as well as the surface absorptivity, are strongly dependant on temperature.
Only one value of those properties can be used in an analytical method. Kou et al. [35]
addressed the issue of thermal properties varying depending on temperature. They
conducted a theoretical and experimental study of heat-flow and solid-state phase
transformations during the laser surface hardening of 1018 steel with a continuous wave
CO2 laser of 15 kW capacity. They established a three dimensional model using the finite
difference method that takes into account the temperature dependence of the thermal
properties.
Patwa and Shin [17] conducted an accurate study on the modeling of the laser
hardening process of laser operating parameters and initial microstructure without the need
for experimental data. Their model combined a three dimensional transient numerical
solution for an AISI 5150H steel rotating cylinder mounted on a computer-numerically
controlled lathe. They used a kinetic model to determine the phase change using the CCT
diagram of the steel. Their model was able to accurately predict the case depth as well as
the hardness.
Concerning complex geometries such as gears, an analytical study of the laser surface
transformation hardening process was conducted by R. Komanduri and Z.B. Hou [36] to
determine the temperature distribution generated by a laser heat source (and thus the depth
and width of the case hardening) depending on the process parameters. They made a
comparative study between their analytical approach and the semi-empirical relationships
developed by Steen and Courtney [34]. Some years later, R. Komanduri and Z.B. Hou [37]
used their established analytical model to optimize the process parameters for different heat
intensity distributions, such as normal (pseudo-Gaussian), uniform, or bimodal (TEM11),
that allow more uniform case hardening depth than the Gaussian distribution (TEM00).
Their study established the theoretical optimized ranges of power and scanning velocity
depending on the heat intensity distribution and the laser beam diameter.
79
An interesting study was conducted by H. Zhang [9] et al. that compared the contact
fatigue strength of carbon case hardening and laser hardening of gear teeth. The carbon
case hardening has the shortcoming of generating very large distortions necessitating
regrinding. The laser hardening process does not have this inconvenience. According to
their study, the contact fatigue strength limit of laser hardened gears was found to be 92 %
that of those treated by carbon case hardening.
The objective of this paper is to present the simulation of the laser surface
transformation in the case of complex geometries, such as gears, which are rarely treated in
the literature. It is an extension of the study conducted by G. Billaud et al [38].The aim is to
develop a method that enables predicting the results of gear treatment without conducting
expensive physical experiments. The simulation is validated by experimental tests using a
test bench made for that purpose. The gears are made of a commonly used AISI 4340 low
alloy steel that is well known in industry for its good hardenability. The proposed approach
examines laser hardening parameters commonly known to influence the characteristics of
hardened surfaces, such as input power, scanning velocity and focus radius. However, the
process presented in this study involves a rotational movement of the gear, so an additional
parameter, the rotation speed, is also taken into account.
3.2.4-Model description
One issue with the laser surface transformation of a rotating workpiece with a lateral
scanning velocity is that the uniformity of treatment is not guaranteed. Indeed, as shown in
Figure 3.1, points A and B, which are diametrically opposed, are not heated by the laser in
the same way.
In practice, in the case of gears, it would mean that two teeth diametrically opposed
would not be heated and treated the same way. To avoid this problem, it is best to have a
great rotation speed and a slow scanning velocity. However, if the scanning velocity is too
slow, there is a risk that the surface temperatures may reach the melting temperature of the
steel, which is undesirable. On the other hand, if the rotation speed is too high, the gear
80
teeth might not be heated enough to create a proper case depth. A compromise must
therefore be found between these two parameters.
Laser
trajectory
Vsc
Wr
A
B
Figure 3.1 : Laser trajectory
In this study, the gear is made of AISI 4340 low alloy steel with 48 teeth. The gear
has an external diameter of 105 mm (from the top of a tooth to the top of the opposite
tooth), an internal diameter of 30 mm, and a width of 7 mm. In order to simulate the
heating process using the finite element method, the part was built on a computer aided
design software.
The gear is mounted on a shaft and clamped by two mounting rings on each side, as
shown in Figure 3.2.
Figure 3.2 : The gear mounted on the shaft and clamped by two mounting rings
81
3.2.4.1-Heat equation
In this study, the heat flux used for the simulation was considered as a Gaussian type
beam distribution, which is given by equation 1 [2].
0
2 2
0
2E E exp{-2 [(y ( )) ] / }sc focy V t D
(1)
E0 and α are given by equation 2 and 3.
2
0 2 P 1 Rc / (π )focE D
(2)
sin cosr rw t z w t x (3)
𝐷𝑓𝑜𝑐 is the Gaussian beam diameter, P is the laser beam power, Rc is the reflection
coefficient of the surface of the chosen material [3]. 𝑉𝑠𝑐 is the scanning velocity, y0 is the
beam center coordinate along the y-axis at t = 0 s. E (in W/m2) represents a Gaussian heat
flux moving along the y-axis with a velocity 𝑉𝑠𝑐 and simultaneously rotating around that
axis with a rotation speed wr.
Finally, α is orthonormal to the laser beam direction at the surface of the workpiece.
Table 3.1: Material properties
Property Symbol Unit Value
Reflection coefficient Rc 0.6
Eutectoid temperature Ac1 K 996
Austenitization temperature Ac3 K 1053
Austenite grain size (assumed) g m 10
Activation energy of carbon diffusion in ferrite Q kJ/mol 80
Pre-exponential for diffusion of carbon D0 m2/s 6x10-5
Gas constant R J/mol.K 8.314
Steel carbon content C 0.43 %
Austenite carbon content Ce 0.8 %
Ferrite carbon content Cf 0.01 %
Critical value of carbon content Cc 0.05 %
Volume fraction of pearlite colonies fi 0.5375
82
As will be shown later in this study, the process requires a preheating of the gear
teeth in order to create a decent, rather uniform, case depth. Therefore, in this study, T0 will
be the initial temperature of the gear teeth, and that temperature will be the chosen
preheating temperature. The preheating experimental procedure is explained later in the
paper.
Figure 3.3 shows the isothermal contours (zoomed in on few teeth) for a power of
2500 W, a scanning velocity of 1 mm/s and a rotation speed of 240 rpm. The preheating
temperature (initial temperature) T0 is set at 873 K.
Figure 3.3 : Isothermal contours for P = 2500 W, Vsc = 1 mm/s, wr = 240 rpm,
T0 = 873 K
Figure 3.3 shows that the temperature is at its maximum at the location of the laser,
and greatly decreases both with the depth and after the laser has passed. The rapid decrease
of temperature is essential for quenching of the steel to occur. However, in this figure, it
can be seen that the surface temperature is too high (about 1871 K), and above the melting
temperature of AISI 4340 steel (about 1450 K). An experiment under these conditions
would most likely lead to a partial melting of the gear teeth, which is not desirable. It is
worth noting that there is a heat concentration at the top of the teeth since the gear has sharp
Time = 2.308161 s Isosurface : Temperature (K) Arrow Volume : Total
heat flux
83
corners. Concentrating the heat at such points could cause slight melting that would bow
them during heat treatment.
Figure 3.4, Figure 3.5 and Figure 3.6 show the isothermal contours resulting from
the simulation using the same laser power and scanning velocity at the same time step, but
with different rotation speeds and a lower preheating temperature of 500 K. The figures
show that the rotation plays a major role in the uniformity of the treatment. High rotation
speeds greatly reduce the peak temperature (providing the other control parameters remain
the same). At 60 rpm, the temperature increases to 1560 K (at t = 2.306089 s). At 120 rpm
it increases to 1448 K at the same time step, while at 300 rpm it only increases to 957 K,
which is not enough to achieve a proper case depth. Indeed, in the last case the peak
temperature is under the austenitization temperature.
Figure 3.4 : Isothermal contours for P = 1500 W, Vsc = 1 mm/s, wr = 60
rpm, T0 = 500 K
Time = 2.306089 s Isosurface : Temperature (K) Arrow Volume : Total heat flux
84
Figure 3.5 : Isothermal contours for P = 1500 W, Vsc = 1 mm/s, wr
= 120 rpm, T0 = 500 K
Figure 3.6 : Isothermal contours for P = 1500 W, Vsc = 1 mm/s, wr =
300 rpm, T0 = 500 K
Time = 2.306089 s Isosurface: Temperature (K) Arrow Volume : Total heat flux
85
3.2.4.2-Metallurgical equations
Once the temperature distribution is generated by the software, the temperature data
can be extracted to determine the metallurgical transformations that occur during the
surface transformation. The following metallurgical equations for a low alloy steel,
established by Ashby and Easterling [5] are used to calculate the case depth.
The total number of diffusive jumps which occurs during the heat cycle affects the
extent of the structural change and is given by the kinetic strength [5].
I = exp -Q/ R×T t dt (4)
Q is the activation energy for the transformation and R is the gas constant. The
kinetic strength can also be simply expressed as in equation 5.
pI = α τ×exp -Q/ R×T (5)
Tp is the peak temperature at the considered depth and τ is the thermal time
constant. The term α and τ are approximated by the equations 6 and 7.
pα = 3 R×T /Q (6)
1
p 0τ = 1-Rc P/ 2 π K e V T -T (7)
T0 represents the initial preheating temperature.
The obtained austenite has the same carbon content as a perlite microstructure ce =
0.8 %. From there, the carbon diffuses in the proeutectoid ferrite. When the temperature
reaches the austenitization temperature (about 1000 K), the volume fraction of austenite is
the volume fraction fi (which is also the minimum volume fraction of martensite).
i f ff = C-C / 0.8-C C/0.8
(8)
Time = 2.306089 s Isosurface: Temperature (K) Arrow Volume : Total heat flux
86
cf is the negligible carbon content of the ferrite and c is the carbon content of the
steel.
The maximum martensite fraction allowed by the transformation temperature time
diagram (TTT diagram) is
p 1
i p 1 3 1 1 p 3
fm = 0 if T < Ac
fm = fi+ 1-f T -Ac / Ac -Ac if
®®®®®®
Ac T Ac
p 3 ®®®®®® fm = 1 if T > Ac
(9)
Ac1 is the austenitization temperature (the start of the transformation). Ac3 (about
1073 K) is the temperature of complete austenitization (Table 3.1).
Ashby and Easterling supposed that all the material with a specific carbon
proportion above a critical value, Cc, will transform into martensite. The volume fraction of
the martensite is then given by equation 10 [2, 5].
2/3
i i e c 0f = fm- fm-f ×exp - 12 f / g π ×ln C / 2 C D ×I (10)
g is the mean grain size and D0 is the diffusion constant for the carbon in ferrite.
The hardness can then be calculated by a mixture rule (equation 11).
m f+pH=f×H + 1-f ×H (11)
The values Hm and Hf+p are given by Maynier equations that take into account the
cooling rate and the composition [21].
3.2.5-Experimental validation
3.2.5.1-Experimental setup
The experimental validation is performed by using a commercial 3 kW Nd: Yag
laser (IPG YLS-3000-ST2), combined with an articulated robot with 6 degrees of freedom
87
(Figure 3.7). This type of laser generates a beam with a wavelength λ = 1064 µm. The
gear is mounted on a lathe protected by a casing. The protective casing is fixed on a table
and is placed under the laser head. The process parameters are the input power, the
scanning velocity and the lathe’s rotation speed. In this study, the laser beam has a
continuous straight-line trajectory back and forth.
Figure 3.7 shows the complete experimental setup with the protective casing
closed, the laser head above and ready to fire, and a thermal camera in the foreground to
monitor the rising temperature.
Figure 3.7 : Experimental Setup.
3.2.5.2-Preheating step
The preheating procedure consists of a series of very high speed laser sweeps back
and forth in a short time and at moderately low power while the gear mounted on lathe is
rotating.
A thermal camera indicates the temperature in the gear teeth in real time. The
selected power for the preheating procedure depends on the desired preheating temperature.
88
However, the duration of the preheating treatment, which depends on the number of sweeps
and their velocity, is always kept the same.
3.2.5.3-Heating treatment
Immediately after the preheating process, a surface transformation is performed by a
single low speed laser scanning along the width of the gear at a high power. The gear is still
rotating at the same speed during that step. The gear teeth quench themselves by rapid heat
conduction into the colder core of the gear. The mounting rings and the shaft are heat
conductors that facilitate rapid cooling.
3.2.5.4-Experimental tests and discussion
With the help of the simulation, a series of experimental tests was conducted in
order to determine the effect of each parameter on the case depth at the top, the bottom and
on the slope of the gear teeth, along the entire width of the gears.
Figure 3.8 and Figure 3.9 show the treatment for two different scanning velocities
(the other process parameters and the preheating procedure remain the same). The small
treated zone appears darken under the surface of the tooth at the top, the bottom and on the
slope of the tooth. The core of the material (in bright) remained unaffected by the
treatment.
89
Figure 3.8 : Case depth observation for P = 3000 W, Vsc = 0.75
mm/s, wr = 300 rpm and T0 = 500 K
In the Figure 3.10it can be seen that the treatment was rather uniform at the top, the
bottom and on the slope of the tooth, with only a slightly deeper case depth at the top.
However, in this case, the case depth is very shallow (no more than 200-300 µm) and might
not be sufficient to significantly improve material properties such as wear resistance. By
comparison, gears treated by induction show much greater case depths on their teeth.
However, the drawback of the induction process is that it requires very high power (tens of
kilowatts) to perform a decent heat treatment.[39]
In Figure 3.9, the scanning velocity was decreased and, as expected, the treated
zone is shallower than that shown in Figure 3.8. Indeed, a slower scanning velocity means
that there will be more energy brought into the part. Thus, the case depth increases when
the scanning velocity decreases. However the case depth is still greater at the top of the
tooth than it is on the slope and at the bottom. In addition, it can be seen in Figure 3.9 than
the shape of the tooth was slightly bowed, due to the fact that the teeth originally had sharp
corners that concentrated the heat. The simulation also demonstrated this, as can be seen in
90
Figure 3.3 where the temperature is higher in the sharp corners at the top of the tooth. This
slight deformation is tolerable though considering the dimensions of the teeth.
Figure 3.9 : Case depth observation for P = 3000 W, Vsc= 0.50
mm/s and wr = 300 rpm and T0 = 500 K
The simulation showed that the depth of the treatment largely depends on the
preheating treatment. The closer the preheating temperature comes to the austenitization
temperature, the deeper more the treated zone.
Figure 3.10 shows that all the teeth of the gears are roughly treated the same way
with the same case depth observed on consecutive teeth.
91
Figure 3.10 : Case depth observation for P = 2500 W, Vsc = 1
mm/s and wr = 240 rpm and T0 = 600 K
3.2.5.5-Comparison between simulation and experiment
Three tests are conducted to study the effect of the power and the rotation velocity
on the case depth, as well as to validate the simulated hardness profile (see Table 3.2).
The scanning velocity and the preheating temperature, T0, remain the same (0.75
mm/s and 600 K, respectively) for each of the three tests.
Figure 3.11, 3.12 and 3.13 show the hardness profiles obtained by both the
simulation and the experiment according to the values presented in Table 3.2. In both the
simulation and the experiment, the hardness profile is measured on the head of a tooth at its
center.
Table 3.2 : Experimental matrix
Test Power (W) Rotation Speed (rpm)
1 2500 240
2 2500 300
3 3000 300
92
Figure 3.11 : Hardness curve for P = 2500 W, Vsc = 0.75 mm/s
and wr = 240 rpm and T0 = 600 K
The figures clearly show three distinct zones: the treated zone, with a high hardness
and a complete transformation into martensite; the transition zone, with a decreasing
hardness where a partial transformation occurred; and the unaffected zone, where the
hardness did not change. No transformation happened in the unaffected zone as the
temperature did not reach the austenitization temperature.
The figures also show that the simulated hardness profile fairly closely follows the
experimental hardness profile in each case. This means that with the adjusted 3D model, it
is possible to predict the hardness profile for a given set of process parameters without
conducting actual experiments, which could prove economically advantageous.
As expected, and as predicted by the simulation, Figure 3.12 shows that the case
depth decreases as the rotation speed increases (from 0.3-0.4 µm at 240 rpm to 0.2 µm at
300 rpm).
350
400
450
500
550
600
650
700
750
0 100 200 300 400 500 600 700 800 900 1000
Har
dnes
s (H
V)
Depth (µm)
Simulated hardness (HV)
Experimental hardness (HV)
93
Figure 3.12 : Hardness curves for P = 2500 W, Vsc = 0.75 mm/s
and wr = 300 rpm and T0 = 600 K
Figure 3.13 shows that the case depth increases as the power increases (from 0.2 µm with
2500 W to 0.3 µm with 3000 W).
350
400
450
500
550
600
650
700
0 100 200 300 400 500 600 700 800 900 1000
Har
dnes
s (H
V)
Depth (µm)
Simulated hardness (HV)
Experimental hardness (HV)
94
Figure 3.13 : Hardness curves for P = 3000 W, Vsc = 0.75 mm/s
and wr = 300 rpm and T0 = 600 K
Table 3.3 shows the maximum absolute error and relative error, for each of the three
tests, between the simulated hardness and the experimental harness. Because the reflection
coefficient of the surface is hard to accurately estimate and has an effect on the case depth,
there are some small differences between the simulated hardness profile and the
experimental hardness profile. Indeed, a greater reflection coefficient, for example, means
that less energy will be absorbed by the part. In this study, the reflection coefficient was
assumed to be 0.6 (which is a common reflection coefficient for mild steels [3]), but it
might be slightly different. In any case, using a surface coating could be a way to obtain a
better estimation of the reflection coefficient.
Table 3.3 : Maximum absolute and relative hardness errors
Test Absolute error (HV) Relative error (%)
1 40 9.45
2 48 10.28
3 57 10
The maximum relative error for the three tests was 10.28 %, which indicates a
decent model accuracy.
350
400
450
500
550
600
650
700
0 100 200 300 400 500 600 700 800 900 1000
Har
dnes
s (H
V)
Depth (µm)
Simulated hardness (HV)
Experimental hardness (HV)
95
In the three previous tests, the hardness profile was measured on the top of a tooth.
Indeed, as is shown in Figure 3.9 and Figure 3.10, the base of the teeth is always less
treated than the top. It appears than the chosen parameters for the three tests did not permit
the formation of a significant case depth at the base of the gear teeth.
Moreover, the low rotation speeds (240 rpm or 300 rpm) induced a partial bowing
of the tops of the teeth, as was shown in Figure 3.9 and Figure 3.10. Increasing the
rotation speed or decreasing the power or the scanning velocity to avoid bowing is useless,
as this would result in no treatment at all.
In the next experiment, the power used during the preheating step is increased in
order to reach a preheating temperature, T0, of 873 K (the austenitization temperature Ac1
being at 1000 K). The simulation showed that a greater preheating temperature allowed
treatment at greater rotation speeds and lower power, and without the surface reaching the
melting temperature.
Figure 3.14 : Case depth observation for P= 1500W,
Vsc = 0.75 mm/s and wr = 750 rpm and T0 = 873 K
Figure 3.14 shows a treatment with a power of 1500 W, a rotation speed of 750 rpm, a
scanning velocity of 0.75 mm/s and a preheating temperature, T0, of 873 K.
The figure clearly shows that there is treatment at the base of the teeth.
96
The case depth obtained is very comparable to the case depth obtained by the induction
process [39] but is achieved using a much lower input power (only 1500 W was used in this
study, compared to the tens of thousands of watts usually required to treat gear teeth by
induction). It is also noteworthy that Figure 3.14 shows that the top of the tooth is not
bowed as the melting temperature was not reached for this test. This is especially apparent
in comparing Figure 3.9 and Figure 3.14.
The case depth achieved at the top of the gear with a high preheating temperature is
much deeper than that seen using a low preheating temperature.
Figure 3.15 : Hardness curves for P= 1500 W, Vsc = 0.75 mm/s
and wr = 750 rpm and T0 = 873 K at the top of the tooth
Figure 3.15 and Figure 3.16 show the simulated and the predicted hardness in the
case of a preheating temperature of T0 = 873 K. The case depth at the top of the teeth
exceeds 1 mm, while the case depth at the base of the teeth is about 0.2-0.3 mm. Once
again, it appears that the base of the gear teeth is difficult to properly treat compared to the
top.
300
350
400
450
500
550
600
650
700
750
0 100 200 300 400 500 600 700 800 900 1000
Har
dnes
s (H
V)
Depth (µm)
Simulated hardness (HV)
Experimental hardness (HV)
97
Once again, the simulation was able to fairly closely match the experiment, as it is
shown by Table 3.4, which gives the maximum absolute and relative errors for the top and
the bottom of the gear teeth.
Table 3.4 : Comparison between simulation and experiment for T0 = 873 K
Location Absolute error (HV) Relative error (%)
Top of the tooth 43 6.41
Foot of the tooth 40 8.33
Figure 3.16 : Hardness curves for P = 1500W, Vsc = 0.75 mm/s
and wr = 750 rpm and T0 = 873 K at the foot of the tooth
3.2.6-Conclusion
The Thermal analysis of surface transformation hardening of gears using laser is
conducted using a 3D numerical model validated by an experimental procedure. A 3D
model of the laser hardening process in the case of gear teeth is implemented on a
commercial software capable of solving the governing heat flow equation using the finite
element method. By extracting the temperature data, it is possible to produce a simulated
hardness profile using the Ashby and Easterling equations. The simulation proves to be
quite accurate with a maximum relative error around 10 %. The observations made with the
300
350
400
450
500
550
600
650
700
750
0 100 200 300 400 500 600 700 800 900 1000
Har
dnes
s (H
V)
Depth (µm)
Simulated hardness (HV)
Experimental hardness (HV)
98
simulation are easily verified by the experiment. The case depth increases as the power
increases and/or as the rotation speed decreases. However, due to the geometry of the gear
teeth, low rotation speeds with high laser power generate partial melting of the tips of the
teeth. The simulation also shows that the preheating temperature plays a major role in the
quality of the treatment. Indeed, with a higher preheating temperature, it is possible to
obtain a very decent case depth at both the top and the base of the teeth. This is validated
by the experiment where a case depth comparable with what can be obtained by induction
is produced. From the results obtained in this study, it appears that it would be possible to
develop an industrial process to treat gear teeth by laser using relatively low power and
rotation speeds. Optimizing the preheating process used in this study in order to reach a
good preheating temperature seems to be a solution. An interesting investigation would be
to test the ultimate bending strength as well as the wear resistance of gear teeth that have
been properly treated by laser.
3.2.7-Acknowledgment
The authors would like to thank AlKhader Borki, research assistant at the University
of Québec at Rimouski, for his involvement in the experimental setup and procedure
99
CONCLUSION GÉNÉRALE
Cette étude a montré qu’il est possible actuellement de modéliser numériquement et
avec une bonne précision le procédé de traitement thermique superficiel au laser des
matériaux à l’aide d’un logiciel de calculs par la méthode des éléments finis. Bien que la
modélisation soit longue à mettre en œuvre à cause du nombre important de paramètres, le
modèle s’avère étonnamment précis et robuste démontrant une concordance remarquable
entre les prédictions et les mesures expérimentales.
Dès lors, une prédiction numérique du résultat en fonction des paramètres d’entrée
devient envisageable et permet d’éviter le long et coûteux processus essai-erreur. La
première phase du projet a permis d’établir un premier modèle numérique fonctionnel du
procédé dans le cas simple d’un parallélépipède rectangle avec une source de chaleur en
translation. Dans cette étude, un logiciel de calculs par élément fini fut utilisé pour résoudre
l’équation de propagation de la chaleur dans le matériau avec des paramètres dépendant de
la température. L’historique de température fut ensuite extrait pour établir les profils de
dureté — dureté en fonction de la profondeur — grâce aux équations métallurgiques. La
vérification expérimentale a donné de bons résultats avec des profils de dureté prédits très
proches des profils de dureté réels mesurés par micro-indentation sous les mêmes
paramètres de contrôle. Cependant, il est vite apparu que la modélisation par une source de
chaleur compliquait la mise en place du modèle (notamment par la difficulté à déterminer le
coefficient d’extinction Ac) ainsi que la résolution de ce dernier. De plus, le modèle serait
difficilement adaptable à des géométries plus complexes.
C’est la raison pour laquelle il a été décidé de changer le mode dans le deuxième
chapitre et de modéliser le laser par un flux de chaleur à travers la surface du matériau au
lieu d’une source de chaleur. Cette modélisation nous a permis d’éviter le coefficient
d’extinction problématique ainsi que de préparer le passage aux géométries plus complexes
par une approche plus robuste. Dans ce deuxième chapitre, la géométrie utilisée est la
même que pour le premier chapitre avec le même mouvement de translation. Toutefois,
quelques tests ont été nécessaires afin d’ajuster le coefficient de réflexion de matériau qui
semblait dépendant de la température en surface et donc des paramètres de contrôle.
100
Comme dans le premier chapitre, les résultats expérimentaux ont vite montré une
concordance avec les résultats donnés par la simulation. Cependant, la modélisation par
flux de chaleur a permis de diminuer drastiquement le temps de calcul et donc de générer
un grand nombre de données. Le grand nombre de données a permis de générer des
formules de régression très précises reliant la profondeur et la largeur durcie aux
paramètres de contrôle du procédé. Le réseau de neurones artificiels, une fois entrainé,
pouvait fonctionner indépendamment de tous logiciels de calculs par éléments finis et
donnait des résultats très proches de la réalité pourvu que les paramètres d’entrées fussent
inclus dans sa plage d’entrainement. Un tel réseau de neurones artificiels pourrait
éventuellement servir à l’élaboration de logiciels avec des applications industrielles dans le
domaine du traitement thermique au laser.
Ce modèle par flux de chaleur a permis un passage relativement facile aux
géométries complexes dans le troisième chapitre. Dans ce troisième chapitre, un
mouvement de rotation a été rajouté en plus du mouvement de translation. Le modèle a
permis de simuler le traitement au laser complet des dents d’un engrenage monté sur un
tour. Pour traiter convenablement les dents d’un engrenage sans causer de déformations
thermiques, il est apparu qu’un préchauffage à basse puissance avec un balayage en va-et-
vient est nécessaire afin d’approcher la température des dents le plus près possible de la
température d’austénitisation. Le traitement lui-même consiste ensuite en un seul balayage
à forte puissance. Les résultats expérimentaux ont démontré que les résultats donnés par la
simulation étaient proches de la réalité avec des profils de dureté simulés et réels presque
identique. La simulation a aussi montré que, par ce procédé, le sommet de la dent est
toujours traité plus en profondeur que le pied. Cette observation a ensuite été confirmée
plus tard par les essais expérimentaux. En revanche, à cause de la géométrie complexe et du
mouvement de rotation, le temps de simulation a drastiquement augmenté par rapport à
celui observé dans le deuxième chapitre. Il est toujours possible de générer un grand
nombre de données et de les utiliser pour réaliser des études statistiques ou entrainer un
réseau de neurones artificiels capable de prédire les profondeurs et largeurs traitées pour
des paramètres d’entrées donnés, mais le procédé est significativement plus long. Par
101
ailleurs, le banc d’essai fabriqué pour la validation expérimentale a permis d’établir un
procédé de traitement des dents d’engrenages par laser donnant des résultats relativement
identiques à ceux obtenus par le traitement par induction. Ce procédé requiert des niveaux
puissances très inférieurs à ceux mis en jeu dans le procédé par induction. Il a aussi
l’avantage d’être adaptable à n’importe quel diamètre d’engrenage contrairement au
procédé par induction qui requiert un équipement spécifique pour chaque diamètre
d’engrenage.
D’une façon générale, les modèles développés ont permis de prédire assez
convenablement les résultats d’un traitement thermique superficiel au laser, pour des
ensembles de paramètres de contrôle donnés, sans recourir à la coûteuse méthode consistant
à avancer par échecs/corrections avec des essais expérimentaux. Une piste intéressante
serait d’utiliser ces modèles afin de générer un grand nombre de données pour entrainer des
réseaux de neurones bien plus perfectionnés que celui du deuxième chapitre. Ces réseaux
de neurones pourraient servir à créer des logiciels que les industriels utiliseraient pour
économiser en temps et en coût. Aussi, le procédé de traitement au laser d’un engrenage
monté sur un tour développé dans le troisième chapitre pourrait avoir de possibles
applications industrielles s’il était davantage approfondi.
RÉFÉRENCES BIBLIOGRAPHIQUES
[1] Ion JC. Laser processing of engineering materials : principles, procedure and industrial
application. Amsterdam: Elsevier/Butterworth-Heinemann; 2005.
[2] Kannatey-Asibu E. Principles of laser materials processing. Hoboken, N.J.: Wiley;
2009.
[3] Steen WM, Mazumber J. Laser Material Processing. 4th ed. ed. London: Springer;
2010.
[4] Chipman J. Thermodynamics and phase diagram of the Fe-C system. MT. 1972;3:55-
64.
[5] Ashby MF, Easterling KE. The transformation hardening of steel surfaces by laser
beams—I. Hypo-eutectoid steels. Acta Metallurgica. 1984;32:1935-48.
[6] Krauss G. Steels : heat treatment and processing principles. Materials Park, Ohio: ASM
International; 1993.
[7] Paul E. Materials and Processes in Manufacturing (8th edition): Macmillan; 1997.
[8] Shiue RK, Chen C. Laser transformation hardening of tempered 4340 steel. MTA.
1992;23:163-70.
[9] Zhang H, Shi Y, Xu C, Kutsuna M. Comparison of contact fatigue strength of carbon
case hardening and laser hardening of gears. Surface engineering. 2004;20:117-20.
[10] America LIo, Ready JF, Farson DF. LIA handbook of laser materials processing. 1st
ed. ed. Orlando, Fla. : Magnolia Publishing: Laser Institute of America; 2001.
[11] Totten GE. Handbook of residual stress and deformation of steel: ASM international;
2002.
[12] Steen WM. Laser Material Processing. 4th ed.. ed. London: Springer; 2010.
[13] Davis M, Kapadia P, Dowden J, Steen WM, Courtney CHG. Heat hardening of metal
surfaces with a scanning laser beam. Journal of Physics D: Applied Physics. 1986;19:1981.
[14] Mioković T, Schulze V, Vöhringer O, Löhe D. Prediction of phase transformations
during laser surface hardening of AISI 4140 including the effects of inhomogeneous
austenite formation. Materials Science and Engineering: A. 2006;435–436:547-55.
[15] Skvarenina S, Shin YC. Predictive modeling and experimental results for laser
hardening of AISI 1536 steel with complex geometric features by a high power diode laser.
Surface and Coatings Technology. 2006;201:2256-69.
[16] Tobar MJ, Álvarez C, Amado JM, Ramil A, Saavedra E, Yáñez A. Laser
transformation hardening of a tool steel: Simulation-based parameter optimization and
experimental results. Surface and Coatings Technology. 2006;200:6362-7.
[17] Patwa R, Shin YC. Predictive modeling of laser hardening of AISI5150H steels.
International Journal of Machine Tools and Manufacture. 2007;47:307-20.
103
[18] Orazi L, Fortunato A, Cuccolini G, Tani G. An efficient model for laser surface
hardening of hypo-eutectoid steels. Applied Surface Science. 2010;256:1913-9.
[19] Antony J, Jiju Antony F. Teaching the Taguchi method to industrial engineers. Work
Study. 2001;50:141-9.
[20] Badkar D, Pandey K, Buvanashekaran G. Parameter optimization of laser
transformation hardening by using Taguchi method and utility concept. Int J Adv Manuf
Technol. 2011;52:1067-77.
[21] Doane D. Application of hardenability concepts in heat treatment of steel. Journal of
Heat Treating. 1979;1:5-30.
[22] Yang ST, Matthews MJ, Elhadj S, Cooke D, Guss GM, Draggoo VG, et al. Comparing
the use of mid-infrared versus far-infrared lasers for mitigating damage growth on fused
silica. Appl Opt. 2010;49:2606-16.
[23] McDaniels RL, White SA, Liaw K, Chen L, McCay MH, Liaw PK. Effects of a laser
surface processing induced heat-affected zone on the fatigue behavior of AISI 4340 steel.
Materials Science and Engineering: A. 2008;485:500-7.
[24] Chen SL, Shen D. Optimisation and Quantitative Evaluation of the Qualities for Nd-
YAG Laser Transformation Hardening. Int J Adv Manuf Technol. 1999;15:70-8.
[25] Benyounis KY, Olabi AG, Hashmi MSJ. Effect of laser welding parameters on the
heat input and weld-bead profile. Journal of Materials Processing Technology. 2005;164–
165:978-85.
[26] Sathiya P, Abdul Jaleel MY, Katherasan D. Optimization of welding parameters for
laser bead-on-plate welding using Taguchi method. Prod Eng Res Devel. 2010;4:465-76.
[27] Hagan MT, Demuth HB, Beale MH. Neural network design: Pws Boston; 1996.
[28] Ciurana J, Arias G, Ozel T. Neural Network Modeling and Particle Swarm
Optimization (PSO) of Process Parameters in Pulsed Laser Micromachining of Hardened
AISI H13 Steel. Materials & Manufacturing Processes. 2009;24:358-68.
[29] Pan QY, Huang WD, Song RG, Zhou YH, Zhang GL. The improvement of localized
corrosion resistance in sensitized stainless steel by laser surface remelting. Surface and
Coatings Technology. 1998;102:245-55.
[30] Munteanu A. SURFACE HARDNESS PREDICTION USING ARTIFICIAL
NEURAL NETWORKS IN CASE OF ELECTRON BEAM MACHNING PROCESS.
2012.
[31] Haykin S. Neural networks: a comprehensive foundation, 1994. Mc Millan, New
Jersey. 2010.
[32] Principe J, Euliano N, Lefebvre W. Neural and adaptive systems: fundamentals
through simulations. 2000. John Wiley and Sons, New York, pgs.119:514.
[33] Munikamal T, Sundarraj S. Modeling the Case Hardening of Automotive Components.
Metall and Materi Trans B. 2013;44:436-46.
[34] Steen W, Courtney C. Surface heat treatment of EnS steel using a 2kW continuous-
wave CO2 laser. Metals Technology. 1979;6:456-62.
[35] Kou S, Sun DK, Le YP. A Fundamental Study of Laser Transformation Hardening.
MTA. 1983;14:643-53.
[36] Komanduri R, Hou Z. Thermal analysis of the laser surface transformation hardening
process. International Journal of heat and mass transfer. 2001;44:2845-62.
104
[37] Komanduri R, Hou Z. Thermal analysis of laser surface transformation hardening—
optimization of process parameters. International Journal of Machine Tools and
Manufacture. 2004;44:991-1008.
[38] Billaud G, El Ouafi A, Barka N. ANN Based Model for Estimation of Transformation
Hardening of AISI 4340 Steel Plate Heat-Treated by Laser. Materials Sciences and
Applications. 2015;6:978.
[39] Barka N, Chebak A, El Ouafi A, Jahazi M, Menou A. A New Approach in Optimizing
the Induction Heating Process Using Flux Concentrators: Application to 4340 Steel Spur
Gear. J of Materi Eng and Perform. 2014;23:3092-9.